The state of the art in kidney and kidney tumor segmentation in contrast-enhanced CT imaging: Results of the KiTS19 Challenge

Abstract

There is a large body of literature linking anatomic and geometric characteristics of kidney tumors to perioperative and oncologic outcomes. Semantic segmentation of these tumors and their host kidneys is a promising tool for quantitatively characterizing these lesions, but its adoption is limited due to the manual effort required to produce high-quality 3D segmentations of these structures. Recently, methods based on deep learning have shown excellent results in automatic 3D segmentation, but they require large datasets for training, and there remains little consensus on which methods perform best. The 2019 Kidney and Kidney Tumor Segmentation challenge (KiTS19) was a competition held in conjunction with the 2019 International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) which sought to address these issues and stimulate progress on this automatic segmentation problem.A training set of 210 cross sectional CT images with kidney tumors was publicly released with corresponding semantic segmentation masks. 106 teams from five continents used this data to develop automated systems to predict the true segmentation masks on a test set of 90 CT images for which the corresponding ground truth segmentations were kept private. These predictions were scored and ranked according to their average Sørensen-Dice coefficient between the kidney and tumor across all 90 cases. The winning team achieved a Dice of 0.974 for kidney and 0.851 for tumor, approaching the inter-annotator performance on kidney (0.983) but falling short on tumor (0.923). This challenge has now entered an "open leaderboard" phase where it serves as a challenging benchmark in 3D semantic segmentation.

Kidney segmentation in neck-to-knee body MRI of 40,000 UK Biobank participants

“…In terms of methodology, kidney segmentations with Dice scores of up to 0.974 have been reported in the literature for benchmark challenges involving neural networks on CT data. 13,18 Reaching comparable quality on the UK Biobank neck-toknee body MRI may not be technically feasible, as the given images are of lower resolution and even repeat segmentation by human operators yielded lower consistency in this work. With no fixed image contrast, such as the Hounsfield units in CT, an objectively consistent placement of the kidney outline in MRI is more challenging.…”

“…For the kidney in particular, various approaches have been proposed predominantly for image data from Computed Tomography (CT), including techniques based on statistical shape models and region growing 10 , graph cuts 11 , and deformable boundaries 12 . Contemporary benchmark challenges are increasingly dominated by machine learning techniques such as deep learning with convolutional neural networks, as seen in the MICCAI 2019 Kidney and Kidney Tumor Segmentation (KiTS19) 13 and with CT image data, in which similar approaches have also been proposed for measurements of total volume in subjects with ADPKD. 14 Fully convolutional networks for semantic image segmentation 15 range from architectures such as the U-Net with 2D data 16 to 2.5D 17 and 3D techniques, 18 which are able to learn the task of segmenting specific image structures from reference data in training.…”

MixCL: Pixel label matters to contrastive learning

“…Datasets Up to 765 CT scans with annotations are used for pre-training, which come from (1) Medical Segmentation Decathlon (MSD) dataset [1] (only Liver, Lung and Pancreas are used for pre-training), (2) NIH Pancreas-CT [16], and (3) 2019 Kidney Tumor Segmentation Challenge (KiTS) [10]. The pre-trained model is then fine-tuned on two datasets: The Beyond the Cranial Vault (BTCV) 3 and Spleen segmentation in MSD.…”

MixSearch: Searching for Domain Generalized Medical Image Segmentation Architectures

“…The performance of different models generated by various search methods is evaluated using Dice score and Jaccard score [47] on three datasets, including ISIC [41], CVC [42] and CHAOS-CT [43], which are collected from dermoscopy, gastroscopy and CT equipments, respectively. In order to evaluate the generalization ability of learned architecture across different datasets and domains, we also conduct extensive experiments on three additional datasets (unseen in search stage) which are ETIS [48], KiTS [49] and LiTS [50].…”

“…In order to test the generalization ability of learned network architectures, we conduct experiments on three unseen datasets: ETIS [48], LiTS [50] and KiTS [49], which come from different modalities and domains. ETIS is collected using wireless capsule endoscopy equipments, while LiTS and KiTS contain CT slices for the liver and kidney, respectively.…”