Complete List of Challenge Submissions
An attempt at beating the 3D U-Net
Isensee, Fabian
Maier-Hein, Klaus H.
The U-Net is arguably the most successful segmentation architecture in the medical domain. Here we apply a 3D U-Net to the 2019 Kidney and Kidney Tumor Segmentation Challenge and attempt to improve upon it by augmenting it with residual and pre-activation residual blocks. Cross-validation results on the training cases suggest only very minor, barely measurable improvements. Due to marginally higher dice scores, the residual 3D U-Net is chosen for test set prediction. With a Composite Dice score of 91.23 on the test set, our method outperformed all 105 competing teams and won the KiTS2019 challenge by a small margin.
Cascaded Semantic Segmentation for Kidney and Tumor
Hou, Xiaoshuai
Xie, Chunmei
Li, Fengyi
Yang, Nan
Automated detection and segmentation of kidney tumors from 3D CT images is very useful for doctors to make diagnosis and treatment plan. In this paper, we described a multi-stage semantic segmentation pipeline for kidney and tumor segmentation from 3D CT images based on 3D U-Net architecture. The current method can achieve 0.9XX, 0.8XX average dice for kidney and tumor in the KiTS19 challenge.
Segmentation of kidney tumor by multi-resolution VB-nets
Mu, Guangrui
Lin, Zhiyong
Han, Miaofei
Yao, Guang
Gao, Yaozong
Accurate segmentation of kidney tumors can assist doctors to diagnose diseases, and to improve treatment planning, which is highly demanded in the clinical practice. In this work, we propose multi-resolution 3D V-Net networks to automatically segment kidney and renal tumor in computed tomography (CT) images. Specifically, we adopt two resolutions and propose a customized V-Net model called VB-Net for both resolutions. The VB-Net model in the coarse resolution can robustly localize the organs, while the VB-Net model in the fine resolution can accurately refine the boundary of each organ or lesion. We experiment in the KiTS19 challenge, which shows promising performance.
Cascaded Volumetric Convolutional Network for Kidney Tumor Segmentation from CT volumes
Zhang, Yao
Wang, Yixin
Hou, Feng
Yang, Jiawei
Xiong, Guangwei
Tian, Jiang
Zhong, Cheng
Automated segmentation of kidney and tumor from 3D CT scans is necessary for the diagnosis, monitoring, and treatment planning of the disease. In this paper, we describe a two-stage framework for kidney and tumor segmentation based on 3D fully convolutional network (FCN). The first stage preliminarily locate the kidney and cut off the irrelevant background to reduce class imbalance and computation cost. Then the second stage precisely segment the kidney and tumor on the cropped patch. The proposed method achieves 98.05% and 83.70% of Dice score on the validation set of MICCAI 2019 KiTS Challenge.
Solution to the Kidney Tumor Segmentation Challenge 2019
Ma, Jun
Kidney and kidney tumor segmentation are essential steps in kidney cancer surgery. In this paper, we focus on addressing hard cases and exploring the kidney tumor shape prior rather than developing new convolution neural network architectures. Specifically, we train additional tumor segmentation networks to bias the ensemble classifier to tumor. Moreover, we propose the compact loss function to constrain the shape of the tumor segmentation results. Experiments on KiTS challenge show that both hard mining and compact can improve the performance of U-Net baseline.
Coarse to Fine Framework for Kidney Tumor Segmentation
Liu, Shuolin
Accurate segmentation of kidney tumor is a key step in image-guided radiation therapy. However, shapes, scales and appearance vary greatly from patient to patient, which pose a serious challenge to segment targets correctly. In this work, we proposed a coarse-to-fine framework to automatically segment kidney and tumor computed tomography (CT) images. Specifically, we adopt two resolutions and propose a improved 3D U-Net network for kidney tumor segmentation. The model in the coarse resolution can robustly localize the kidney, while the model in the fine resolution can accurately refine the boundary of kidney and tumor.
Multi Scale Supervised 3D U-Net for Kidney and Tumor Segmentation
Zhao, Wenshuai
Zeng, Zengfeng
U-Net has achieved huge success in various medical image segmentation challenges. Kinds of new architectures with bells and whistles might succeed in certain dataset when employed with optimal hyperparameter, but their generalization always can’t be guaranteed. Here, we focused on the basic U-Net architecture and proposed a multi scale supervised 3D U-Net for the segmentation task in KiTS19 challenge. To enhance the performance, our work can be summarized as three folds: first, we used multi scale supervision in the decoder pathway, which could encourage the network to predict right results from the deep layers; second, with the aim to alleviate the bad effect from the sample imbalance of kidney and tumor, we adopted exponential logarithmic loss; third, a connectedcomponent based post processing method was designed to remove the obviously wrong voxels. In the published KiTS19 training dataset (totally 210 patients), we divided 42 patients to be test dataset and finally obtained DICE scores of 0.969 and 0.805 for the kidney and tumor respectively. In the challenge, we finally achieved the 7th place among 106 teams with the Composite Dice of 0.8961, namely 0.9741 for kidney and 0.8181 for tumor.
Fully Automatic Segmentation of Kidney and Tumor Based on Cascaded U-Nets
Li, Yu
Fully automatic segmentation of kidney and its lesions is an important step to obtain accurate clinical diagnosis and computer aided decision support system. In this paper, a method of automatic segmentation of kidney and renal tumor in CT abdominal images using cascade 3D U-Net convolutional neural network (3D cU-Nets) is presented. We trained and cascaded two 3D U-Nets for the joint segmentation of kidney and renal tumor. In the first step, we trained a 3D U-Net to segment kidney as the ROI input for the second 3D u-net.The second 3D U-Net only segmented the lesion from the renal ROI predicted in step 1.
Edge-Aware Network for Kidneys and Kidney Tumor Semantic Segmentation
Myronenko, Andriy
Hatamizadeh, Ali
Automated segmentation of kidneys and kidney tumors is an important step in quantifying the tumor’s morphometrical details to monitor the progression of the disease and accurately compare decisions regarding the kidney tumor treatment. Manual delineation techniques are often tedious, error-prone and require expert knowledge for creating unambiguous representation of kidneys and kidney tumors segmentation. In this work, we propose an end-to-end boundary aware fully Convolutional Neural Networks (CNNs) for reliable kidney and kidney tumor semantic segmentation from arterial phase abdominal 3D CT scans. We propose a segmentation network entailing an encoder-decoder that specifically accounts for organ/tumor edge information by devising a dedicated network edge branch and edge-aware loss terms. We have evaluated our model on 2019 MICCAI KiTS Kidney Tumor Segmentation challenge dataset. Based on our own data split, we achieved 0.970 dice for kidney & tumor, and 0.834 dice for tumor segmentation.
Segmentation of CT Kidney and kidney tumor by MDD-Net
Chen, Ball
Accurate segmentation of kidney and kidney tumor is an important step for treatment. Due to the wide variety in kidney and kidney tumor morphology, it’s really a challenging task. In this work, we propose the Multi-level double-dimension Network to automatically segmentat kidney and kidney tumor. We select the modified FPN as backbone and aggregate different scale information from multi levels to make the final prediction. In the KiTS 2019, we use 170 CT scans for training and the remaining 40 CT scans are used to evaluate the model. At the time of submission, we obtained the best result by averaging multiple models.
Ensemble U-Net for 2019 Kidney Tumor Segmentation Challenge
Wu, Ting
Its known to us all that convolutional network makes medical processing more accurate and efficient as a significant tool for assisting doctors. In deed,aiming at kidney and diversity of kidney tumor,there already have various effective segmentation results from networks learning, and they are more comparable. Therefore,methods based on networks has become a mainstream in image processing.For this MICCAI kidney and kidney tumor segmentation challenges, we proposed our own scheme.We are inspired by U-Net,experiment in five U-Net on 300 abdominal CT scan of arterial phase in patients with renal cell carcinoma, then take all results as an ensemble and use it as the final result.
Segmentation of Kidney and Renal Tumor in CT Scans Using Convolutional Networks
Yuan, Shaofeng
Yang, Feng
Tang, Yujiao
Xing, Yanyan
Zhang, Liyun
Accurate segmentation of kidney and renal tumor in CT images is a prerequisite step in surgery planning. However, this task remains a challenge. In this report, we use convolutional networks (ConvNet) to automatically segment kidney and renal tumor. Specifically, we adopt a 2D ConvNet to select a range of slices to be segmented in the inference phase for accelerating segmentation, while a 3D ConvNet is trained to segment regions of interest in the above narrow range. In localization phase, CT images from several publicly available datasets were used for learning localizer. This localizer aims to filter out slices impossible containing kidney and renal tumor, and it was fine-tuned from AlexNet pre-trained on ImageNet. In segmentation phase, a simple U-net with large patch size (160×160×80) was trained to delineate contours of kidney and renal tumor. In the 2019 MICCAI Kidney Tumor Segmentation (KiTS19) Challenge, 5-fold cross-validation was performed on the training set. 168 (80%) CT scans were used for training and remaining 42 (20%) cases were used for validation. The resulting average Dice similarity coefficients are 0.9662 and 0.7905 for kidney and renal tumor, respectively.
BiSC-UNet: A fine segmentation framework for kidney and renal tumor
Wang, Chuanxia
He, Yuting
Qi, Xiaoming
Zhao, Ziteng
Yang, Guanyu
Zhu, Xiaomei
Zhang, Shaobo
Dillenseger, Jean-Louis
Coatrieux, Jean-Louis
Computed tomography (CT) images can provide a view of the patients’ internal organs. This property is particular conducive to surgery planning. As one of the most common cancers, renal cancer also can be treated effectively by the laparo-scopic partial nephrectomy (LPN). However, automatic and accurate kidney and renal tumor segmentation in CT images remains a challenge. In this paper, BiSC-UNet framework which combines two different resolution SC-UNets is proposed for kidney and renal tumor fine segmentation. Rough SC-UNet is in charge of locating the kidney and renal tumor roughly to achieve the kidney region of interest (ROI) in original CT images. The other fine SC-UNet utilizes the kidney ROI for final fine kidney and renal tumor segmentation. In the proposed SC-UNet, not only the labeled kidney and renal tumor are used for training, but also the labeled renal tumor edge and kidney edge are utilized for accurate segmentation result. A balanced cross-entropy which selects and dynamic weights different regions and categorys is proposed for category imbalance. Extensive experiments on CT datasets demonstrate the effectiveness of proposed network.
KiTS challenge: VNet with attention gates and deep supervision
Tureckova, Alzbeta
Turecek, Tomas
Kominkova, Zuzana
Rodŕıguez-Sánchez, Antonio
This paper presents the 3D fully convolutional neural network extended by attention gates and deep supervision layers. The model is able to automatically segment the kidney and kidney-tumor from arterial phase abdominal computed tomography (CT) scans. It was trained on the dataset proposed by the Kidney Tumor Segmentation Challange 2019. The best solution reaches the dice score 96, 43± 1, 06 and 79, 94± 5, 33 for kidney and kidney-tumor labels, respectively. The implementation of the proposed methodology using PyTorch is publicly available at github.com/tureckova/Abdomen-CT-Image-Segmentation.
Applying nnU-Net to the KiTS19 Grand Challenge
Graham-Knight, J. B.
Djavadifar, A.
Lasserre, P.
Najjaran, H.
U-Net, conceived in 2015, is making a resurgence in medical semantic segmentation tasks. This comeback is largely thanks to the excellent performance of nnU-Net in recent competitions. nnU-Net generalizes well, as proven by its first-place finish in the Medical Segmentation Decathalon. Notably, nnU-Net focuses on the training process rather than algorithmic improvements, and can often beat more complex algorithms. This paper shows the results of applying nnU-Net to the KiTS19 Kidney Segmentation Grand Challenge. Each of the 5 cross-validation training folds achieves good results, with scores nearing or exceeding 0.9 after approximately 500 epochs per fold.
3D Unet-based Kidney and Kidney Tumer Segmentation with Attentive Feature Learning
Tao, Qingyi
Wu, Zhonghua
Cheong, Isaac
Yang, Jingyi
Ge, Zongyuan
Lin, Guosheng
Cai, Jianfei
To study the kidney diseases and kidney tumor from Computed Tomography(CT) imaging data, it is helpful to segment the region of interest through computer aided auto-segmentation tool. In the KiTs 2019 challenge [1], we are provided 3D volumetric CT data to train a model for kidney and kidney tumor segmentation. We introduce an improved deep 3D Unet by enriching the feature representation in CT images using an attention module. We achieve 1.5% improvement in the segmentation accuracy when evaluated on the validation set.
Triplanar 3D-to-2D networks with dense connections and dilated convolutions: application to the KITS 2019 challenge
McKinley, Richard
Wiest, Roland
von Tengg-Kobligk, Hendrik
We describe a method for the segmentation of kidney and kidney tumors based on computed tomography imaging, based on the KITS 2019 challenge dataset.
Automatic Kidney and Tumor Segmentation with Hybrid Hierarchical Networks
Yuan, Yading
Automatic segmentation of kidney and its tumors is an essential but challenging step for extracting quantitative imaging biomarkers for accurate tumor detection, diagnosis, prognosis and treatment assessment. Kidney Tumor Segmentation Challenge (KiTS) provides a common platform for comparing different automatic algorithms on abdominal CT images in tasks of 1) kidney segmentation and 2) kidney tumor segmentation . We participate this challenge by developing a fully automatic framework based on deep neural networks. By observing that clinicians usually contour organs and tumors in the axial view while evaluating the contours in 3D space, we adopt a 3-step hierarchical structure with hybrid 2D and 3D models. In the first step, a simple 2D U-Net model is trained to obtain a quick but coarse segmentation of the kidney region on the entire 3D CT volume; then another 2D U-Net using residual blocks with channel-wise attention is applied to each kidney region for kidney and tumor segmentation. At last, the segmented tumor is refined by a 3D model for final tumor segmentation. Our framework was trained using the 210 challenge training cases provided by KiTS. By 5-fold evaluation, our method achieved an average Dice Similarity Coefficient (DSC) of 0.970 on kidneys and 0.756 on kidney tumors, respectively.
A Coarse-to-fine Framework for Automated Kidney and Kidney Tumor Segmentation from Volumetric CT Images
Chen, Huai
Wu, Xiyi
Huang, Yijie
Wang, Lisheng
Automatic semantic segmentation of kidney and kidney tumor is a promising tool for the treatment of kidney cancer. Due to the wide variety in kidney and kidney tumor morphology, it is still a great challenge to complete accurate segmentation of kidney and kidney tumor. We propose a new framework based on our previous work accepted by MICCAI2019, which is a coarse-to-fine segmentation framework to realize accurate and fast segmentation of kidney and kidney tumor.
Technical report KiTS 2019
Zakirov, Adel
This technical report on my solution for KiTS 2019 Challenge. I build an end-to-end pipeline for kidney and tumor segmentation. My pipeline consists of three modules: (1) coarse segmentation step, (2) fine kidney segmentation, (3) fine tumor candidates segmentation on the kidney(s), (4) choosing between candidates. All 4 steps are based on 3D sem-seg model with some modifications in their architecture, training policy, data representation.
KiTS19 Challenge Segmentation
Tsai, Yi-Chin
Sun, Yung-Nien
There are more than 400,000 new cases of kidney cancer each year, and surgery is its most common treatment. Due to the wide variety in kidney and kidney tumor morphology, there is currently great interest in how tumor morphology relates to surgical outcomes, as well as in developing advanced surgical planning techniques. Automatic semantic segmentation is a promising tool for these efforts, but morphological heterogeneity makes it a difficult problem. The goal of this challenge is to accelerate the development of reliable kidney and kidney tumor semantic segmentation methodologies. We have produced ground truth semantic segmentations for arterial phase abdominal CT scans of 300 unique kidney cancer patients who underwent partial or radical nephrectomy at our institution. 210 of these have been released for model training and validation, and the remaining 90 will be held out for objective model evaluation. Recently, fully convolutional neural networks (FCNs), including 2D and 3D FCNs, serve as the backbone in many volumetric image segmentation. However, 2D convolutions cannot fully leverage the spatial information along the third dimension while 3D convolutions suffer from high computational cost and GPU memory consumption. Our method consists of 2.5D convolutions for efficiently extracting g intra-slice and inter-slice features and densely supervised, data augmentation for generate better segmentation.
SERU: cascaded SE-ResNeXT U-Net for kidney and tumor segmentation on KITS2019
Li, Lei
Lian, Sheng
Luo, Zhiming
Accurate segmentation of kidney tumor in CT images is a challenging task. For solving this, we proposed SE-ResNeXT U-Net (SERU) model, which combines the advantages of SE-Net, ResNeXT and U-Net. For utilizing context information and key slices’ information, we implement our model in a coarse-to-fine manner. We find left and right kidney’s key slice respectively, and obtain key patches for refine training. We train and test our method on the KiTS19 Challenge. The predictions on kidney segmentation and tumor segmentation by our model show promising results.
Kidney tumor segmentation using an ensembling multi-stage deep learning approach. A contribution to the KiTS19 challenge
Santini, Gianmarco
Moreau, Noémie
Rubeaux, Mathieu
Precise characterization of the kidney and kidney tumor characteristics is of outmost importance in the context of kidney cancer treatment, especially for nephron sparing surgery which requires a precise localization of the tissues to be removed. The need for accurate and automatic delineation tools is at the origin of the KiTS19 challenge. It aims at accelerating the research and development in this field to aid prognosis and treatment planning by providing a characterized dataset of 300 CT scans to be segmented. To address the challenge, we proposed an automatic, multi-stage, 2.5D deep learning-based segmentation approach based on Residual UNet framework. An ensambling operation is added at the end to combine prediction results from previous stages reducing the variance between single models. The results obtained are promising and could be improved by incorporating prior knowledge about the benign cysts that regularly lower the tumor segmentation results.
Multi-Encoder U-Net for Automatic Kidney Tumor Segmentation
Chen, Xueying
Xu, Chao
Kidney tumor segmentation is a difficult yet critical task for medical image analysis. In recent years, deep learning based methods have achieved many excellent performances in the field of medical image segmentation. In this paper, we propose a Multi-Encoder U-Net segmentation method to tackle the challenging problem of kidney tumor segmentation from CT images. Our Multi-Encoder U-Net method uses three different depth networks as encoders for kidney tumor segmentation: VGG16, ResNet34, ResNet50, a feature fusion networkFED-Net is also used simultaneously, finally fusing the four results. We tested our method on the dataset of MICCAI 2019 Kidney Tumor Segmentation Challenge(KiTS).
Two stages kidney and tumor segmentation(kits2019)
Yiwen, Zhang
In order to accurately segment kidney tumors, this paper uses a twostage segmentation/ Stage one performs a coarse step positioning of the kidney. Stage two performs an accurate segmentation of the kidneys and tumors.
Kidney Segmentation Framework using 3D CNN
Meng, Zhe
Two 3D CNN was used to detect the location of kidney and separate kidney and tumor respectively. 1 Network structure The CNN structure we used is V-net [1]. The left side of the network is divided in different stages that operate at different resolutions. Each stage comprises one to three convolutional layers. The convolutions performed in each stage use volumetric kernels having size 5 × 5 × 5 voxels. As the data proceeds through different stages along the compression path, its resolution is reduced. This is performed through convolution with 2× 2× 2 voxels wide kernels applied with stride 2 . Since the second operation extracts features by considering only non overlapping 2 × 2 × 2 volume patches, the size of the resulting feature maps is halved. 2 Kidney detection One V-net was training using patches extracted from whole CT images. In testing stage, it was used firstly detect where are the kidney. 3 Kidney and tumor segmentation Another V-net was training using patches extracted around kidneys. In testing stage, it was used to separate kidney and tumor based on previous detection. References 1. Fausto Milletari1,: V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation.
Two-Stage Method for Kidney and Tumor Segmentation Based On Cascade Res-VNet
Hao, Xiaoyu
In this work, We proposed a two-stage method for segmentation of kidneys and kidney tumors in CT images based on cascade Res-VNet. In the first stage, we consider the kidney and tumor as a whole region and use a cascade Res-VNet to do the segmentation. In the second stage, we firstly extract patches of ROIs based on the results of the first stage and use some tricks to improve precision. Then, we use another cascade Res-VNet to segment the tumor. Ouput from each stage are combined together as the final results.
KiTS19 Submission
Kondo, Satoshi
This report describes our method submitted to 2019 Kidney Tumor Segmentation (KiTS19) Challenge. Our method employs two step approach. In the first step, an input volume is divided into two-dimensional images in three orthogonal planes and the two-dimensional images are fed into encoder-decoder networks to segment kidney and tumor regions. In the second step, the segmentation results for three orthogonal planes are fed into convolutional neural networks to obtain final segmentation results. Although our method is based on twodimensional segmentation, but three-dimensional like processing can be performed by combining segmentation results in three orthogonal planes.
Kits19TumorSegmentation with VNet
Chen, Junqiang
we design the deep learning network Vnet for segmentation tumor and kidney.Fist, preprocess the kidney and kidney tumor data,second, segmentating kidney progress split into two steps: Corse segmentation and fine segmentation.third,segmentation kidney tumor process split into two steps: 2d segmentation and 3d segmentation
A three stage segmentation method of kidney and kidney tumor
Wei, Zhan
The morphometry of a kidney tumor revealed by contrast enhanced Computed Tomography (CT) imaging is an important factor in clinical decision making surrounding the lesion’s diagnosis and treatment. I Use 3D Unet to segment kidney and kidney tumor respectively.
Kidney Tumour Segmentation
Sharma, Rochan
Medical Image Segmentation is a challenging field in the area of Computer Vision. In this work Two deep learning models were explored namely U-Net and ENet. The reason to shortlist U-Net was it is suitable on a small data set and also originally designed for Biomedical Image segmentation. However when compared to ENet it is much slower. To speed up the process of Kidney Tumor segmentation , ENet was shortlisted and also experimented on the data set provided. ENet was very fast as compared to U-Net , However some visual representations of the predicted results have shown promising results in U-Net better then ENet. A classification model called as Xception Model was also considered right in the first phase , so as to shortlist those slices from the CT which have the presence of Kidney. So that the output obtained can be given as an input to semantic segmentation model. This would allows us to speed up the process of Kidney Tumor Segmentation. Finally performance parameter which was used for evaluation segmentation models was IOU.
3D Segmentation For Kidney Data
Cui, Zhiying
The segmentation pipeline consists two parts, first, we process 3d data into a series of 2d images, then send them into a popular segmentation neural network-deeplab v3+ to get segmentation result.
SUBMANIFOLD SPARSE CONVOLUTIONAL NETWORKS
Retinskiy, Dmitry
We present method for effective kidney and kidney’s tumor segmentation based on the 3-dimensional model reconstructed from the computed tomography (CT) input images. Analysis of the 3D model is done using submanifold sparse convolutional network (SSCN).
Kidney tumor segmentation using a 2D U-Net followed by a statistical post-processing filter
Paolucci, Iwan
Each year, there are about 400’000 new cases of kidney cancer worldwide causing around 175’000 deaths. For clinical decision making it is important to understand the morphometry of the tumor, which involves the timeconsuming task of delineating tumor and kidney in 3D CT images. Automatic segmentation could be an important tool for clinicians and researchers to also study the correlations between tumor morphometry and clinical outcomes. We present a segmentation method which combines the popular U-Net convolutional neural network architecture with post-processing based on statistical constraints of the available training data. The full implementation, based on PyTorch, and the trained weights can be found on GitHub http://github.com/ipa/kits2019.
Kidney and Tumor Segmentation Based on 3D Context Extracting
Yan, Bin
Organ segmentation and lesion detection play a vital role in the computer-aided diagnosis (CAD) systems. The task of this Kits challenge is about kidney and tumor segmentation. We proposed an effective model to complete this Kits challenge. Our model receives part of body 3D scans as input, and outputs the probability map of the input scans. 2D contexts of intra-slices are extracted by VGG network, and 3D contexts of inter-slices are presented by concatenating the 2D contexts. Then proposals are extracted by region proposal network (RPN), while 3D context are regarded as auxiliary information for region of interest (ROI) regression, classification and mask generation. Our model has shown promising result for this Kits challenge.
Kidney and Kidney-tumor Segmentation Using Cascaded V-Nets
Arafat Hussain, Mohammad
Kidney cancer is the seventh most common cancer worldwide, accounting for an estimated 140,000 global deaths annually. Kidney segmentation in volumetric medical images plays an important role in clinical diagnosis, radiotherapy planning, interventional guidance and patient follow-ups however, to our knowledge, there is no automatic kidneytumor segmentation method present in the literature. In this paper, we address the challenge of simultaneous semantic segmentation of kidney and tumor by adopting a cascaded V-Net framework. The first V-Net in our pipeline produces a region of interest around the probable location of the kidney and tumor, which facilitates the removal of the unwanted region in the CT volume. The second sets of V-Nets are trained separately for the kidney and tumor, which produces the kidney and tumor masks respectively. The final segmentation is achieved by combining the kidney and tumor mask together. Our method is trained and validated on 190 and 20 patients scans, respectively, accesses from 2019 Kidney Tumor Segmentation Challenge database. We achieved a validation accuracy in terms of the Sørensen Dice coefficient of about 97%.
Kidney and tumor segmentation using an ensemble of deep neural networks
Wu, Yu
Gan, Yu
Wu, Yuhang
Yi, Zhang
For the segmentation of kidney and tumor task, we propose a two stages model that consists of several classification networks and segmentation models. The first stage is the foreground and background classification subnetwork, this stage is to recognize whether there are kidneys or tumors on images, so we propose a classification model called RD-Net which can effectively reduce the errors caused by a large of background images and improve the efficiency of the whole segmentation results. The second stage is the segmentation model used to predict the contour of the target (kidney or tumor). Therefore, we propose Att-ResUnet model and multi-scale ensemble of postprocessing methods used to integrate the predicted results of multiple models, so as to improve the accuracy of prediction results.
Kidney Tumor Segmentation Using 3D U-nets for Masking and Labeling
Ito, Yuichi
Nishimoto, Takuya
Chou, Weizhong
Ishiguro, Toshitaka
Mori, Kensaku
Kojo, Kosuke
Kojima, Takahiro
Nakagawa, Tsutomu
Kakeya, Hideki
In this paper we report on the algorithm we applied to KiTS 19 Challenge. We used 3D U-nets for masking and labeling separately. We mainly focused on data augmentation and grouping of training data to improve the result of segmentation.
H-DenseUNet for Kidney and Tumor Segmentation from CT Scans
Li, Xiaomeng
Liu, Lihao
Heng, Pheng-Ann
Automatic kidney tumor segmentation from CT scans is an essential step for computer-aided diagnosis of cancer. In this paper, we present an improved H-DenseUNet for kidney and tumor segmentation. Specifically, we first train the DenseUNet and then fine tune the network with the 3D counterpart. To further increase the performance, we employ both cross-entropy and dice loss. We evaluate our method on the 2019 MICCAI kidney and tumor segmentation challenge. We split the training dataset of the challenge to 200 training set and 10 validation set. On the validation set, our method achieves 97.0% (Dice) for kidney segmentation and 67.2% (Dice) for tumor segmentation. This model is submitted to the challenge for final performance evaluation on the test dataset.
The Method for Kits19 Challenge
Su, Chengwei
Du, Bo
There are more than 400,000 new cases of kidney cancer each year [1], and surgery is its most common treatment [2]. Due to the wide variety in kidney and kidney tumor morphology, there is currently great interest in how tumor morphology relates to surgical outcomes, [3,4] as well as in developing advanced surgical planning techniques [5]. Automatic semantic segmentation is a promising tool for these efforts, but morphological heterogeneity makes it a difficult problem. In this paper, we use ResUNet to solve this problem. The ResUNet combines the UNet with residual connection, which is fast and has less parameters. The source code can be found at: https://github.com/FlyGlider/Kits19
Minimal Information Loss Attention U-Net for abdominal CT of kidney cancers segmentation
Liu, Yubai
Jia, Fucang
Qi, Shouliang
Recent work has shown that U-net is a straight-forward and successful architecture, it quickly evolved to a commonly used benchmark in medical image segmentation, Which nnU-Net had better performance We improved the nnUNet model by incorporating a new image pyramid to preserve contextual features and attention gate. In order to let different kinds of class details more easily accessible at different scales, we injected the encoder layers with an input image pyramid before each of the max-pooling layers. We proposed a new image pyramid mechanism with dilated convolution that counters the loss of information caused by max-pooling, re-introducing the original image at multiple points within the network. We evaluated this model in the 2019 Kidney Tumor Segmentation Challenge. and got the dice coefficient 0.958 of kidney and 0.847 of tumors.
Segmentation of kidney and kidney tumor by cascaded fusion FCNs with soft-boundary regression
Zhang, Jian
He, Kelei
Qin, Tiexin
Chen, Jianrong
Yang, Lihe
To produce reliable kidney and kidney tumor semantic segmentation, we proposed a two-stage method to automatically segment kidney and tumor. Specifically, in the first stage, to crop input into a small region, we train a small network to locate kidney and tumor with down-sampled image. In second stage, we train three types of networks to segment kidney, tumor, kidney and tumor respectively. Then we combine these networks together with ensemble method to produce reliable kidney and tumor segmentation. Our method can achieve an overall approximate score of 85.1% in DSC in Kits19 Challenge, with 96.9% for kidney and 73.3% for kidney tumor.
Segmentation of kidney lesions with attention model based on Deeplab
Wen, AiQing
Chen, Xiaochuan
Chen, Anni
Shi, Hongyu
Hong, Yuming
We participate this challenge by developing a hierarchical framework. We build the model from two fully convolutional networks: (1) a simple Unet model to normalize the input iamges, (2) a segmentaion network which is an attention model based on Deeplab model. Two models are connected in tandem and trained end-to-end. To ensure a better results, we use the preprocess method proposed by nnUnet in our experiments.
3D Cascaded U-Net with a Squeeze-and-Exicitation Block for Semantic Segmentation on Kidney and Renal Cell Carcinoma in Abdonimal Volumetric CT
Ham, Sungwon
Kim, Sungchul
Yun, Jihye
Chang, Yongjin
Yu, Donghoon
Seo, Hyungi
Yoon, MyungKi
Soo Kyung, Yoon
Kim, Namkug
Segmentation is a fundamental process in medical image analysis. Recently, convolutional neural networks (CNNs) has allowed for automatic segmentation; however, segmentaiton of complex organs and diseases including the kidney or renal cell carcinoma (RCC) remains a different task due to limited data and labor-intensive labeling work. The purpose of this study is to segment kideny and RCC in CT using cascaded 3D U-Net with a squeeze-and-excitation (SE) block using a cascaded method. 210 kidneys and their RCC in abdominal CT images were used as training and validation sets. The Dice similarity coefficients (DSCs) of kidney and RCC in test set were 0.963 and 0.734 respectively. The cascaded semantic segmentation can potentially reduce segmentation efforts and increase the efficiency in clinical workflow.
Renal Tumor Segmentation in CT using Cascade U-Net with 2.5D approach
HE, HE
Shun Leung, Ping
Wang, Lu
We propose Cascade U-Net with 2.5D approach to segment kidney and tumor from 3D CT image. We use standard U-Net to generate segmentations per each volume slide (2D image). 4 prediction volumes are generated per different magnification and slice direction. Then, consolidate the volumes to formulate the final prediction volume. Per experiment on the KiTS19 dataset, we get a 12% raise in dice coefficient when compare with single U-Net prediction.
Rediminds Entry in KiTS19 Competition
Alia, Hamid
Zeffiroa, Trevor
Reddy Nallabasannagaria, Anubhav
Reddiboinaa, Madhu
Bhandaria, Mahendera
Since its presentation in 2015, U-Net has been rapidly adopted by the medical imaging community. However, due to the large number of interrelated parameters, different implementations of preprocessing, architechture, and training can result in wildly different levels of performance, thus, establishing a baseline by which we can compare the results of different methods. As such, in our submission we submit the results of high-resolution 3 dimensional segmentation of kidney and tumor using NN-Unet.
Submission to the Kidney Tumor Segmentation Challenge 2019
Yu, Pengxin
Cui, Xing
Tian, Xi
Ma, Jiechao
Zhang, Rongguo
In this report, we present our method description of the submission to Kidney Tumor Segmentation Challenge 2019. In this challenge, the goal is to segment the kidney and kidney tumor from the CT scans. Our method is based on a common neural architecture U-Net variant, while we pay more attention to the preprocessing stage to better understand the kidney data and postprocessing stage to reduce false positives. The experiments and results show that our proposed methods increase the segmentation accuracy compared to the basic model.
Report of KiTS19 competition
Bang, Du
Hongyun, Yu
Xiangshang, Zheng
Wang
Hao, Ying
As an important organ of the urinary system, the kidney focuses on generating urine, purifying the blood, and also maintaining water, electrolytes and acid-base balance. Kidney tumors, as one of the most common tumors, are extremely harmful. Once they are found, surgery is the widespread treatment. Therefore, accurate segmentation of renal tumors is of great significance to surgeons performing renal tumor resection. To this end, this paper proposes an effective method for segmentation of the kidney and its tumor for the KiTS19 competition. Specifically, our method first designs a 3D ResUNet framework to segment the whole kidney, and then develops a 2.5D segmentation network to segment the tumors based on the result of kidney segmentation. After validation, the performance of our method reaches a good level under the given metric.
Automatic segmentation of kidney and liver tumors in CT images
Efremova, Dina B.
Konovalov, Dmitry A.
Siriapisith, Thanongchai
Kusakunniran, Worapan
Haddawy, Peter
Automatic segmentation of hepatic lesions in computed tomography (CT) images is a challenging task to perform due to heterogeneous, diffusive shape of tumors and complex background. To address the problem more and more researchers rely on assistance of deep convolutional neural networks (CNN) with 2D or 3D type architecture that have proven to be effective in a wide range of computer vision tasks, including medical image processing. In this technical report, we carry out research focused on more careful approach to the process of learning rather than on complex architecture of the CNN. We have chosen MICCAI 2017 LiTS dataset for training process and the public 3DIRCADb dataset for validation of our method. The proposed algorithm reached DICE score 78.8% on the 3DIRCADb dataset. The described method was then applied to the 2019 Kidney Tumor Segmentation (KiTS-2019) challenge, where our single submission achieved 96.38% for kidney and 67.38% for tumor Dice scores.
Convolutional Neural Network for Kidney and kidney Tumor segmentation
Kumar Anand, Vikas
Aurangabadkar, Pranav
Khened, Mahendra
Krishnamurthi, Ganapathy
In this work, we have attempted to develop algorithms for automatic segmentation of kidney and kidney tumous from CT images. We have exploited encoder decoder architecture of fully convolutional neural network.Pre-processesing steps involves slice extraction, data standardization and Hounsfield unit windowing. The proposed network has been trained on CT images of kidney and kidney tumors with their ground truth. Weighted combination of focal loss and dice loss has been minimized using Adam as optimizer. Dice coefficient of 94.68% and 94.51% has been achieved for kidney and kidney tumor segmentation respectively.
Automatic segmentation of kidney and kidney tumors using the cascaded dense network combined with cLSTM in CT scan
Kim, Jae-Hun
Yang, Ehwa
Kyo Kim, Chan
In this study, we develop the cascaded deep neural network model for automatic segmentation of the kidney and kidney tumors in CT scans. We used the fully dense network (to extract inner-slice image features) combined with bi-cLSTM (to extract inter-slice image features) for segmentation of kidney and kidney tumors. The whole CT scan is preprocessing for resizing (256*256) and intensity normalization (clipping between -600 and 400 and then normalizing between 0 and 1), and then entered into 1 st neural network (growth factor = 8, and the consecutive slices = 3) for segmentation of kidney. And the output of 1 st network is resized into 96*96, and entered into 2 nd neural network, which is the same architecture of 1 st network excepting growth factor = 16, and the consecutive slices = 5. Our cascaded deep neural network showed the dice scores of 0.932 (train), and 0.884 (validation) for segmentation of kidney and the dice scores of 0.845 (train), and 0.696 (validation) for segmentation of kidney tumors.
2019 Kidney Tumor Segmentation Challenge Method Manuscript
Jiao, MengLei
Liu, Hong
This paper framework in detail for KiTS19, which is the 2019 Kidney Tumor Segmentation Challenge. We adopt two model ResUNetSM and DeepLabV3 plus to segment kidney and tumor respectively. Firstly, we propose a model ResUNetSM to segment kidney, which uses ResNet for encoder, and adopts SELayer and MobileBlock for decoder. ResUNetSM also adopts ASPP and skip-connect structure. To segment tumor region, we adopt DeepLabV3 plus and segment tumor in the 3D ROI region from above kidney segmentation results to reduce noise. Finally, we use 3DCRF and 3D connected component analysis as post-processing to improve the final segmentation results. Our framework gets the 96.31% mean dice for kidney and 81.64% mean dice for tumor on validation set.
Two-phase Framework for Automatic Kidney and Kidney Tumor Segmentation
Wei, Hao
Wang, Qin
Zhao, Weibing
Zhang, Minqing
Yuan, Kun
Li, Zhen
Precise segmentation of kidney and kidney tumor is essential for computer aided diagnosis. Considering the diverse shape, low contrast with surrounding tissues and small tumor volumes, it’s still challenging to segment kidney and kidney tumor accurately. To solve this problem, we proposed a two-phase framework for automatic segmentation of kidney and kidney tumor. In the first phase, the approximate localization of kidney and kidney tumor is predicted by a coarse segmentation network to overcome GPU memory limitation. Taking the coarse prediction from first phase as input, the region of kidney and tumor is cropped and trained in an enhanced two-stage network to achieve a fine-grained segmentation result in the second phase. Besides, our network prediction flows into multiple post-processing steps to remove false positive such as cyst by taking medical prior knowledge into consideration. Data argumentation through registration and ensemble models are used to further enhance performance.
Coarse-to-fine Kidney Segmentation Framework Incorporating with Abnormal Detection and Correction
Zhang, Yue
Wu, Jiong
Chen, Yifan
Wu, Ed X.
Tang, Xiaoying
In this work, we formulate this segmentation problem into two sub-task: 1) kidney segmentation 2) tumor segmentation. In the first task, three 2D CNN are used to separate the kidney(including tumor )region with background (two class segmentation). In the second task, one 3D CNN and one two channel 2D CNN are used to separate detect and refine tumor within a relative smaller image region.
Dense Pyramid Context Encoder-Decoder Network for Kidney Lesion Segmentation
Yu, Zhen
Song, Youyi
Qin, Jing
In this manuscript, an automated solution is presented for the kidney lesion segmentation. The proposed method consists of two-stage learning procedures which generating prediction masks for kidney and lesion respectively. Since we adopt 2D axial images from CT scans as evaluation data, it is critical to extract sufficient contextual information for capturing the objects varied significantly in appearance within different slices. Hence, we redesign an encoderdecoder network for more effective feature representations learning. We evaluate our method on 2019 Kidney Tumor Segmentation Challenge. There are total 210 labeled CT scans released as training and validation data. The source code can be found at: https://github.com/Zakiyi/kits_2019_segmentation_challenge.
Automated Machine Learning algorithm for Kidney and Kidney tumor segmentation in Computed Tomography Scans
Bharadwaj, Kss
Pawar, Vivek
In this report,we have described an automated algorithm for accurate segmentation of kidney and kidney tumor from CT scans. The dataset for this problem was made available online as part of KiTS19 Challenge. Our model uses a 2 stage cascaded Residual Unet architecture. The first network is designed to predict (Kidney + Tumor) regions. The second network predicts segmented tumor regions from the output of first net. As a post processing step, we have designed a statistical metric which calculates the standard deviation of derivatives of centre of mass of predicted masks to filter out false positives.The report contains implementation details along with results on validation set.
Automatic 3D Medical Image Segmentation with Densely-Connected Volumetric ConvNets
Zhu, Qikui
Du, Bo
Yan, Pingkun
The Automated medical image segmentation in 3D medical images play an important role in many clinical applications, such as diagnosis of prostatitis, medical image cancer and enlarged medical image. However, it is still a challenging task due to the complex background, lacking of clear boundary and various shape and texture between the slices. In this paper, we propose a novel 3D convolutional neural network with densely-connected layers to automatically segment the medical image. Compared with other methods, our method has three compelling advantages. First, our model can effectively detect the medical image region in a volume-to-volume manner by utilizing the 3D convolution rather than the 2D convolution, which can fully exploit both spatial and region information. Second, the proposed network architecture alleviates the vanishing-gradient problem, strengthens the information propagation between layers, overcomes the problem of overfitting and makes the network deeper by adopting a densely-connected manner. Third, besides the densely-connected manner inside each block, we also adopt the long connections strategy between blocks.
Using Two-stage Network to Segment Kidneys and Kidney Tumors
Chen, Pan
Xu, Chenghai
He, Jie
Sun, Chengwei
Ma, Yingying
Sun, Fenglong
There are many new cases of kidney cancer each year, and surgery is the most common treatment. To assist doctors in surgical planning, an accurate and automatic kidney and tumor segmentation method is helpful in the clinical practice. In this paper, we propose a deep learning framework for the segmentation of kidneys and tumors in abdominal CT images. The key idea is using a two-stage strategy. First, for each case, we use a 3d U-shape convolution network to get the localization of each kidney. Then using next 3d U-shape convolution network we obtain the precise segmentation results of each kidney. Finally, merge the results to obtain the complete segmentation. Also, we try some tricks to improve the performance.
Arkansas AI-Campus Method for the 2019 Kidney Tumor Segmentation Challenge
Causey, Jason L.
Stubblefield, Jonathan
Yoshino, Tomonori
Torrico, Alejandro
Qualls, Jake A
Huang, Xiuzhen
Our Arkansas AI-Campus team participants the 2019 Kidney Tumor Segmentation Challenge (KiTS19) during the past 4 months. Here the paper provides a summary of our methods and validation results for this grand challenge in biomedical imaging analysis.
KidNet: An Automated Framework for Renal Lesions Detection and Segmentation in CT Images
Vesal, Sulaiman
Ravikumar, Nishant
Maier, Andreas K.
Renal lesions segmentation and morphological assessment are essential for improving diagnosis and our understanding of renal cancer, which in turn is imperative for reducing the risk of mortality and morbidity in patients. In this paper, we propose an automatic image based method to first detect kidneys in CT images and then segment both kidneys and lesions in higher resolution. Kidneys are detected using an encoder-decoder method trained on low-resolution images. Based on probability maps generated by detector model, we can identify corresponding kidney regions and segment both kidneys and lesions in higher resolution with reducing the false positive voxels. We evaluate our approach on KITS 2019 challenge data set and demonstrate that our proposed method generalizes to unseen clinical CTs of the abdominal.
Segmentation of the kidney and its tumor
Zhu, Kequan
Shen, Caomin
Peng, Yaxin
Ding, Xiaofeng
At present, computer-aided diagnosis and treatment has become a hot research direction. The segmentation of 3D medical images is an important part of computer-aided diagnosis and treatment. This paper uses a two-stage approach to achieve segmentation of kidney and kidney tumors from 3D CT image.
Two Stream Auto-encoder Decoder Network for Kidney and Tumor Segmentation
Moradi, Pouria
Azad, Reza
Asadi-Aghbolaghi, Maryam
In this competition, we apply two-streams auto-encoder decoder structure for learning kidney and kidneys tumor segmentation. To do so, first, we extract axial layers of the tissues along with their segmentation mask from the 3D volume. These axial layers are then clipped using Hanford distance between +512 to -512 to eliminate non-object of interest. These axial layers are then normalized to form the 2D grayscale images. For each of these normalized images, we generate kidney and kidney tumor masks to train two-stream deep networks. The two-streams deep model learns kidney and tumor masks separately and they generate final mask by concatenating the generated masks. We utilize BCDU-net (extended version of U-Net model) as a deep auto-encoder decoder model for segmentation. We utilize 70% of the Kits19 as the training set and the rest of data as the validation set. Experimental results demonstrate that the proposed structure achieves state-of-the-art performance in the segmentation of kidney and tumor region.
Kidney Tumor Segmentation Based on U-Net and V-Net with Double Loss Function Training
Lv, Yi
Wang, Junchen
Medical image processing plays an increasingly important role in clinical diagnosis and treatment. Using the results of kidney CT image segmentation for three-dimensional reconstruction is an intuitive and accurate method for diagnosis. However, the traditional image segmentation algorithm has poor performance due to the large difference of noise between kidney and CT images, and the manual segmentation by doctors will take a very long time and is inefficient. In this paper, we propose an in-depth learning automatic segmentation method for kidney tumors, including preprocessing of training data, network model used in training process, loss function and post-processing, etc. The results show that the average dice of kidney with tumor was 0.97 in the test set.
A Multi-scale Attention Network for Kidney Tumor Segmentation on CT Scans
Sun, Liyan
Zeng, Weihong
Ding, Xinghao
Huang, Yue
Automatic kidney segmentation is a promising tool for developing advanced surgical planning techniques. However, due to the high morphological heterogeneity within the kidney CT data, the automatic segmentation of kidney and tumor is a difficult problem. Although the state-of-the-art 3D U-Net provides accurate segmentations of medical images, the multi-scale information is underutilized. We propose a multi-scale attention network (MSAN) for automatic kidney tumor segmentation. A multi-scale attention layer is developed to combine the local and global contextual information. Furthermore, a ensemble strategy based on voting mechanism boosts the model performance. We achieve the averaged Dice coefficients 94.83% on kidney and 64.89% on tumor in the validation datasets.
Progressively Training an Enhanced U-Net Model for Segmentation of Kidney Tumors
HE, XueJian
Shun Leung, Ping
Wang, Lu
An enhanced U-Net model with multi-scale inputs and deep supervision are adopted for Kidney tumor segmentation. Focal Tversky Loss is used to train the model, in order to improve the model performance of detecting small tumors. Progressive training is proposed for facilitating model converge. A simple postprocessing method is used to remove segmentation noises. The preliminary results indicate that the proposed model can segment the normal kidney with a satisfactory result; for the tumors with small sizes in low contrast or extreme sizes, there is still a room for improvement.
ResCap: Residual Capsules Network for Medical Image Segmentation
Nguyen, Chanh D.Tr.
Dao, Huu-Hung
Huynh, Minh-Thanh
Phu Ward, Tan
Convolutional neural networks (CNNs) have shown remarkable results for a wide range of task in computer vision. However, CNNs has the limitation of poor translation invariance and lack of information about pose; thus, it requires a lot of data. Capsule networks, however, have the ability to preserve information about the pose. In this paper, we present a capsule-based network for medical image segmentation. We adopt the contracting path of the U-Net architecture. The network achieves the same accuracy as U-Net but is much smaller (0.16% number of parameters compared with U-Net).
Kidney and Kidney Tumor Segmentation Using Two- stage Convolutional Neural Network
Lee, Junghyun
Song, Joonyoung
Yang, Serin
Han, Inhwa
Chul Ye, Jong
Kidney tumor is typically diagnosed using computed tomography (CT) imaging by investigating geometric features of kidney tumor. For a reliable diagnosis and treatment planning, kidney tumor quantification is necessary. However, manual segmentation by human requires time and expertise. In addition, inter/intra variability of segmentation results can lead to suboptimal decision. In this study, we propose the two-stage segmentation method using 2.5D and 3D convolutional neural network for kidney and kidney tumor delineation. The two stage model was trained with multi-task loss for pixel-wise cross-entropy loss function for segmentation task and mean square error function for regression task. Experimental results confirm that the proposed method effectively segments kidney and kidney tumor.
Segmentation of Kidney Tumor
Dahiya, Navdeep
Sharma, Alok
Medical Image Segmentation is a routine task in various clinical settings. There is a great interest in understanding the morphology of different human organs healthy or diseased. Until fairly recently, this task was done manually by clinical experts requiring years of difficult training tailored to specific medical fields. In this article we present a fully autonomous Machine Learning based method to segment Kidney and Tumor from human abdominal CT scans. Manually annotated data (210 datasets) were provided as a part of the Kits2109 MICCAI grandchallenge.
Kidney and Tumor Segmentation Using Modified 3D Mask RCNN
Chen, Cong
Ma, Longfei
Jia, Yan
Zuo, Panli
Detection of kidney tumors and accurate evaluation of their size are crucial for tracking cancer progression. Automating 3D volume detection and segmentation can improve workflow as well as patient care. We adapt the state of the art architecture for 2D object detection and segmentation, Mask RCNN, to handle 3D images and employ it along with U-net to detect and segment kidney and kidney tumor from CT scans. We report on competitive results for the kidney segmentation and kidney tumor segmentation on the 2019 Kidney Tumor Segmentation Challenge data set.
Kits19 segmentation challenge
Sirazitdinov, Ilyas
Ibragimov, Bulat
In this work we briefly describe our methodology for the kidney segmentation problem.
Kits2019 Challeges: Brief Descriptions of the Algorithm and Process
Wu, Songxiong
Bin, Huang
Abstract. Chronic kidney disease affects many people around the world. Computed tomography (CT) is a widely used imaging modality for kidney disease diagnosis and quantification. However, automatic pathological kidney segmentation is still a challenging task due to large variations in contrast phase, scanning range, pathology, and position in the abdomen, etc. In this work, we propose to combine different Window wide window position as a multi-channel input and Unet, for robust kidney or kidney tumors segmentation.
Attention Guided 3D U-Net for KiTS19
Zhong, Zhusi
Zhang, Zhenxi
Jiao, Zhicheng
We use a two-stage 3d U-Net model to predict the multi channels segmentations from coarse to fine. The second stage is guided by the predictions from the first stage. 1 Method We proposed a two stages method to segment CT image from coarse to fine. The two stages are trained with different learning scope and are assigned with different learning missions. 1.1 Stage 1 – Coarse stage Data preprocess. Firstly, we downscale the training data to a normal shape, in order to make sure the model can take a whole image at once. All the images and segmentations are downscale to 128*128*32 (height*width*depth). The segmentation files are transformed to 3-channels arrays, in which the channels-wise pixel values represent kidneys, tumors and the background (without kidneys and tumors) in order. Training. We train the standard 3D U-Net follow with a softmax layer. While training, we apply some data augmentation to the training data, including normalize, random contrast, random flip and random rotate. We input all the 210 cases training data and train the model to regress the multi-channel segmentations. We apply with pytorch, and the learning rate is 0.1 which divide 0.1 in 300000 epochs and 500000 epochs. We use the Binary Cross Entropy Loss as loss function. Predicting. The 90 cases testing images are preprocessed the same with the training images then input to the trained model. The channel-wise predictions are scaled back the original shape. We take the first 2 channels of the predictions, represent as the segmentation of kidneys and tumors, then package as the .nii.gz files.
2019 Kidney Tumor Segmentation Challenge: Medical Image Segmentation with Two-Stage Process
Chen, Tung-I
Wu, Min-Sheng
Chang, Yu-Cheng
Lin, Jhih-Yuan
Since we are trying to deal with the medical images of real patients, the dataset are usually predominantly composed of ”normal” samples. The target classes only appear in a very small portion of the entire dataset, which leads to the so-called class imbalance problem. Besides, there is only a small percentage of foreground inside the ”abnormal” images. The great majority of background leads the significant detrimental effect on training. In such cases, model tends to focus on learning the dominant classes, leading to the poor prediction of minority class. However, the incorrect classification of pathological images can cause serious consequence in clinical practice.
2.5D U-Net with Dense/Residual Layers and Global Context Blocks for Kidney Tumor Segmentation
Qi, Lin
Liu, Shuangwei
Wei, Ying
To address the segmentation of kidney and kidney tumor segmentation on CT scans, we propose a new 2.5D convolutional neural network that can utilize global context and multi-stage features.
A double cascaded framework based on 3D SEAU-Net for kidney and kidney tumor Segmentation
Cheng, Jianhong
Liu, Jin
Liu, Liangliang
Pan, Yi
Wong, Jianxin
Accurate segmentation of kidney and kidney tumor from CT-volumes is vital to many clinical endpoints, such as differential diagnosis, prognosis and radiation therapy planning. While manual segmentation is subjective and time-consuming, fully automated extraction is quite imperative and challenging due to intrinsic heterogeneity of tumor structures. To address this problem, we propose a double cascaded framework based on 3D SEAU-Net to hierarchically and successively segment the subregions of the target. This double cascaded framework is used to decompose the complex task of multi-class segmentation into two simpler binary segmentation tasks. That is to say, the region of interest (ROI) including kidney and kidney tumor is trained and extracted in the first step, and the pre-trained weights are used as the initial weights of the network that is to segment the kidney tumor in second step. Our proposed network, 3D SEAU-Net, integrates residual network, dilated convolution, squeeze-and-excitation network and attention mechanism to improve segmentation performance. To speed training and improve network generalization, we take advantage of transfer learning (i.e., weight transfer) in the whole training phase. Meanwhile, we use 3D fully connected conditional random field to refine the result in post-processing phase. Eventually, our proposed segmentation method is evaluated on KiTS 2019 dataset and experimental results achieves mean dice scores 93.51% for the whole kidney and tumor, 92.42% for kidney and 74.34% for tumor on the training data.
Kidney Tumor Segmentation Using Dual Fully Convolutional Networks
Han, Bing
Li, Mengzhang
Ye, Yue
Kidney tumor segmentation is a hot topic in medical image analysis. In this work, we propose a concise but tricky method, dual Fully Convolutional Networks(FCN) to segment kidney and its surrounding tumor. First FCN is utilized to coarsely detect location of kidney and tumor, while second FCN is employed for refining results of first FCN. In the 2019 Kidney Tumor Segmentation Challenge(KiTS19), 160 patients’ CT scans were used for training and our best result were obtained by this two-step FCN models trained with Focal loss.
Convolution Neural Network Application in Kidney Tumor Segmentation on CT Images
Hunag, Jianping
Lin, Zefang
In this paper, we propose an novel network model which is similar to V-net and prove its superiority and efficiency in tumor segmentation. And The model of segmentation of Kidney is Dense V-Network [1]. Then we ensemble the results of two networks together to get a final predict result for kidney and tumor. In particularly, we apply a series of method to image preprocessing, which is proved to be effective in improving dice.
Coarse-to-fine kidney and tumor segmentation with fully convolutional networks
Shen, Chen
Wang, Chenglong
Oda, Masahiro
Mori, Kensaku
Segmentation is one of the most important tasks in medical image analysis. With the development of deep leaning, fully convolutional networks (FCNs) have become the dominant approach for this task and their extension to 3D achieved considerable improvements for automated organ segmentation in volumetric imaging data. In this paper we demonstrate a coarse-to-fine segmentation method using FCNs for Kits 2019 challenge.
Localization Network and End-to-End Cascaded U-Nets for Kidney Tumor Segmentation
Vu, Minh H.
Grimbergen, Guus
Simkó, Attila
Nyholm, Tufve
Löfstedt, Tommy
Kidney tumor segmentation emerges as a new frontier of computer vision in medical imaging. This is partly due to its challenging manual annotation and great medical impact. Within the scope of the Kidney Tumor Segmentation Challenge 2019, that is aiming at combined kidney and tumor segmentation, this work proposes a novel combination of 3D U-Nets—collectively denoted TuNet—utilizing the resulting kidney masks for the consecutive tumor segmentation. The proposed method achieves a Sørensen-Dice coefficient score of 0.902 for the kidney, and 0.408 for the tumor segmentation, computed from a five-fold cross-validation on the 210 patients available in the data.
MIScnn: A Framework for Medical Image Segmentation with Convolutional Neural Networks and Deep Learning
Müller, Dominik
Kramer, Frank
The increased availability and usage of modern medical imaging induced a strong need for automatic medical image segmentation. Still, current image segmentation platforms do not provide the required functionalities for plain setup of medical image segmentation pipelines. Already implemented pipelines are commonly standalone software, optimized on a specific public data set. Therefore, this paper introduces the open-source Python library MIScnn. The aim of MIScnn is to provide an intuitive API allowing fast building of medical image segmentation pipelines including data I/O, preprocessing, data augmentation, patch-wise analysis, metrics, a library with state-of-the-art deep learning models and model utilization like training, prediction, as well as fully automatic evaluation (e.g. cross-validation). Similarly, high configurability and multiple open interfaces allow full pipeline customization. Running a cross-validation with MIScnn on the Kidney Tumor Segmentation Challenge 2019 data set (multi-class semantic segmentation with 300 CT scans) resulted into a powerful predictor based on the standard 3D U-Net model. With this experiment, we could show that the MIScnn framework enables researchers to rapidly set up a complete medical image segmentation pipeline by using just a few lines of code. Additionally, MIScnn was used to deploy a Residual 3D U-Net model for participating at the Kidney Tumor Segmentation Challenge 2019. The source code for MIScnn is available in the Git repository: https://github.com/frankkramerlab/MIScnn.
Segmentation of Kidney and Tumor using Auxiliary Information
Usman Akbar, Muhammad
Murino, Vittorio
Sona, Diego
Automatic segmentation of organs and tumors is a prerequisite of many clinical application in radiology. The anatomical variability of organs in the abdomen and especially of tumors makes it difficult for many methods to obtain good segmentations. in this report we present a cascade of two convolutional neural networks allowing to segment an organ followed by the segmentation of a tumor. The advantage of the proposed pipeline is that the preliminary organ segmentation, which is a simpler task, helps the further segmentation of the tumor. The proposed system was evaluated using the KiTS19 challange dataset.
Hyper Vision Net: Kidney Tumor Segmentation Using Coordinate Convolutional Layer and Attention Unit
Mansoor Roomi, S.M.Md.
Sabarinathan, D.
Parisa Beham, M.
KiTs19 challenge paves the way to haste the improvement of solid kidney tumor semantic segmentation methodologies. Accurate segmentation of kidney tumor in computer tomography (CT) images is a challenging task due to the non-uniform motion, similar appearance and various shape. Inspired by this fact, in this manuscript, we present a novel kidney tumor segmentation method using deep learning network termed as Hyper vision Net model. All the existing U-net models are using a modified version of U-net to segment the kidney tumor region. In the proposed architecture, we introduced supervision layers in the decoder part, and it refines even minimal regions in the output. A dataset consists of real arterial phase abdominal CT scans of 300 patients, including 45964 images has been provided from KiTs19 for training and validation of the proposed model. Compared with the state-of-the-art segmentation methods, the results demonstrate the superiority of our approach on training dice value score of 0.9552 and 0.9633 in tumor region and kidney region, respectively.
Semantic Segmentation of Kidney Tumor using Convolutional Neural Networks
Daza, Laura
Gómez, Catalina
Arbeláez, Pablo
We present a fully automatic method for segmentation of kidney tumors in CT volumetric data based on DeepLab v3+, the stateof-the-art model in semantic segmentation in natural images. We adapt the architecture to process medical data and reduce the computational complexity to allow training 3D models. We evaluate our approach on the Kidney Tumor Segmentation Challenge 2019 dataset, and define a validation set to experiment with the model’s parameters. In our validation set, we report a dice score of XX for the kidney class and YY for the tumor class.
Sequence and Spatial Feature Fusion for Kidney and Tumor Segmentation From CT Volumes
Wenzhong, Han
Li, Kang
Morphological heterogeneity makes the precise segmentation of both the tumor and the kidney becomes a problem for diagnosis and quantitative analysis of the next treatment. In this work, we propose an encoder-decoder architecture which consist of recent neural network for downsampling and 3D convolutional neural network for upsampling. Recent neural network can integrate contextual information between slices and 3D convolutional neural network can make use of spatial information of one case.
Kidney Tumor Detection using Attention based U-Net
Rathnayaka, Prabod
Jayasundara, Vinoj
Nawaratne, Rashmika
De Silva, Daswin
Ranasinghe, Weranja
Alahakoon, Damminda
The advancement of deep learning techniques has provoked the potential of using Medical Image Analysis (MIA) for disease detection and prediction in numerous ways. This has been mostly useful in identifying tumours and abnormalities in many organs of the human body. Particularly in kidney diseases, the treatment options such as surgery have largely benefitted by the ability to detect tumours in early stages, thereby shifting towards more efficient methods including conservative nephron procedures. Therefore, to enable the early detection of kidney tumours, we propose a convolutional neural network based U-Net architecture which is able to detect tumours using an attention mechanism. The proposed architecture was evaluated using KiTS19 Challenge dataset that includes a collection of multi-phase CT imaging, segmentation masks, and comprehensive clinical outcomes for 300 patients who underwent nephrectomy for kidney tumours. The outcomes demonstrate the ability of the proposed architecture to distinguish images with tumours in the kidney and support early tumour detection.
Kidney and Kidney Tumor Segmentation using a Logical Ensemble of U-nets with Volumetric Validation
O'Reilly, Jamie A.
Manas Sangworasil, Jamie A.
Matsuura, Takenobu
Automated medical image segmentation is a priority research area for computational methods. In particular, detection of cancerous tumors represents a current challenge in this area with potential for real-world impact. This paper describes a method developed in response to the 2019 Kidney Tumor Segmentation Challenge (KiTS19). Axial computed tomography (CT) scans from 210 kidney cancer patients were used to develop and evaluate this automatic segmentation method based on a logical ensemble of fully-convolutional network (FCN) architectures, followed by volumetric validation. Data was pre-processed using conventional computer vision techniques, thresholding, histogram equalization, morphological operations, centering, zooming and resizing. Three binary FCN segmentation models were trained to classify kidney and tumor (2), and only tumor (1), respectively. Model output images were stacked and volumetrically validated to produce the final segmentation for each patient scan. The average F1 score from kidney and tumor pixel classifications was calculated as 0.6758 using preprocessed images and annotations; although restoring to the original image format reduced this score. It remains to be seen how this compares to other solutions.
Development of a Deep Learning Algorithm for Segmentation of Kidney Tumor Imaging
Raein Hashemi, Seyed
Gershman, Boris
Valtchinov, Vladimir I.
Development of a Deep Learning Algorithm for Segmentation of Kidney Tumor Imaging
Nested 3D Neural Networks For Kidney And Tumor Segmentation
Zavala-Romero, Olmo
Perez Matos, Javier
Alvarez Illan, Ignacio
Stoyanova, Radka
Breto, Adrian L.
Xu, Isaac
Zavala-Hidalgo, Jorge
Romero-Centeno, Rosario
Nested 3D Neural Networks For Kidney And Tumor Segmentation
Automatic segmentation of kidney and kidney tumor based on 3D convolutional neural networks
Qiao, Pengchong
Zhao, Bin
Liu, Zhiyang
Kidney cancer is a huge threat to humans, and the surgery is the most common treatment. For clinicians, knowing the morphology of the kidney and kidney tumor in advance may be helpful for surgery. Automatic segmentation of kidney and kidney tumor is a promising approach for these efforts. In this paper, we proposed a based 3D convolutional neural network to segment kidney and kidney tumor using the data from the 2019 Kidney Tumor Segmentation Challenge.
Automatic Kidney and Tumor Segmentation with Attention-based V-Net
Hu, Yucheng
Deng, Han
Zhou, Yang
Chen, Yimin
Hao, Zhou
Yang, Wanqi
Deep learning, especially Convolutional Neural Networks (CNNs) have been implemented to resolve a variety of both computer vision and medical image analysis problems recently. Among a rather wide range of Segmentation CNNs, V-Net is a relatively popular one, which is also an extended version of U-Net which processes 2D images. In this work, we propose an innovative V-Net with a embeded attention module. Inspired by spatial neural attention for generating pseudo-annotations, we modify the Decoupled attention into 3D version and insert it into the V-Net. This CNN network is trained end-to-end on CT volumes, and able to learn to predict segmentation blocks for a certain case. The definition of “block” will be elaborated in Section 3. Finally, the blocks will be concatenated to create a complete segmentation for a single case.
A 2D U-Net for Automated Kidney and Renal Tumor Segmentation
Chen, Joseph
Jin, Benson
Kidney and renal tumor segmentation are critical aspects to the diagnosis process. However, segmentation is a time consuming and tedious task, especially for volume segmentation. To help with this issue, we test a simple two-dimensional U-Net are architecture for automating the segmentation process for both regions of interests. In doing so, we found that the vanilla U-Net was able to achieve a local tumor-kidney test dice of 0.91 and tumor-only dice of 0.25 and leaderboard scores of 0.85 and 0.22.
Convolutional neural network stacking for medical image segmentation in CT scans
Kloenne, Marie
Niehaus, Sebastian
Lampe, Leonie
Merola, Alberto
Scherf, Nico
Computed tomography (CT) data poses many challenges to medical image segmentation based on convolutional neural networks (CNNs). The main challenges in handling CT scans with CNN are the scale of data (large range of Hounsfield Units) and the processing of the slices. In this paper, we consider a framework, which addresses these demands regarding the data preprocessing, the data augmentation, and the CNN architecture itself. For this purpose, we present a data preprocessing and an augmentation method tailored to CT data. We evaluate and compare different input dimensionalities and two different CNN architectures. One of the architectures is a modified U-Net and the other a modified Mixed-Scale Dense Network (MS-D Net). Thus, we compare dilated convolutions for parallel multi-scale processing to the U-Net approach with traditional scaling operations based on the different input dimensionalities. Finally, we combine a set of 3D modified MS-D Nets and a set of 2D modified U-Nets as a stacked CNN-model to combine the different strengths of both model.
Automatic system for the renal and cancer segmentation in CT images
Les, T
Markiewicz, T
Dziekiewicz, M
Lorent, M
This article presents the concept of a complex system for automatic detection of kidneys and kidney tumors in computed tomography images. An effective treatment of cancer depends on a quick and effective diagnosis. Computer support for medical diagnostics is crucial in effective specialists’ analysis. Automatic and accurate location, together with precise detection of the kidney and/or tumor contour is a demanding task. In this article, authors present a complex system for automatic detection of kidneys and kidney tumors, based on machine learning techniques, using the U-Net network. Convolutional neural network recognition results are then processed in multiple stages, using morphological processing, 3D model analysis, geometric coefficients analysis and region-growth implementation. The results of the system detection were compared to the reference images marked by an expert. The system presented in the article is characterized by a very high efficiency of recognition and segmentation of kidney and tumor areas.
Segmentation of large renal tumors in CT images by the integration of deep neural networks and thresholding
Nasiri, Nasim
Mohagheghi, Saeed
Hossein Foruzan, Amir
To segment the kidney and its large tumors, we combine a deep neural network and thresholding technique. The deep network segments kidney, and its output is used to detect probable renal tumors. We compare the kidney volume with a normal kidney shape. Incomplete shapes are searched for tumors. Using a seed point the center of the tumor cluster is defined. Then, the pixels of a slice is labeled as normal or abnormal. The labeled pixels are post-processed using morphological filters to refine the result. The outcome of the algorithm is the tumor volume.
Kidney and tumor segmentation using combined Deep learning method
Kamkova, Yuliia
Ali Qadir, Hemin
Jakob, Ole
Prasanna Kumar, Rahul
This paper presents our method for automatic segmentation for kidney and tumor as part of the Kidney Tumor Segmentation Challenge (KiTS19). The KiTS19 Challenge had released a dataset of 300 unique kidney cancer patients, with manual annotations done by Climb 4 Kidney Cancer (C4KC). Here we have proposed our new combined cascade deep learning (DL) approach for solving the tasks of the challenge. We used deep learning based detection for localising kidney with the tumor, followed by deep learning based segmentation to create the labels for kidney and tumor locally. Our approach resulted in high recall (96.13) and high Jacquard score (95.4) on the randomly selected 30 volumes that were picked as the validation set.