The Liver Tumor Segmentation Benchmark (LiTS).

Area of Science:

• Medical Image Analysis
• Computer-Aided Diagnosis
• Radiology

Background:

• Liver tumors present a significant diagnostic challenge.
• Accurate segmentation of liver and tumors is crucial for treatment planning and monitoring.
• Existing automated segmentation methods require robust benchmarking.

Purpose of the Study:

• To establish and report the results of the Liver Tumor Segmentation Benchmark (LiTS).
• To evaluate the performance of various liver and liver tumor segmentation algorithms using computed tomography (CT) data.
• To identify the strengths and weaknesses of current segmentation techniques and guide future research.

Main Methods:

• A diverse dataset of 131 CT volumes for training and 70 unseen volumes for testing was curated from seven institutions.
• Seventy-five algorithms were submitted and evaluated across three benchmark events (ISBI 2017, MICCAI 2017, MICCAI 2018).
• Performance was assessed using Dice scores for segmentation and lesion-wise recall for detection.

Main Results:

• No single algorithm achieved top performance for both liver and tumor segmentation across all events.
• The best liver segmentation Dice score was 0.963.
• The best tumor segmentation Dice scores ranged from 0.674 to 0.739, and the best lesion-wise recall for detection ranged from 0.458 to 0.554.
• Top-performing segmentation algorithms did not necessarily excel at tumor detection.

Conclusions:

• The LiTS benchmark reveals variability in algorithm performance for liver and tumor segmentation and detection.
• Further research is needed to improve automated liver tumor segmentation and detection accuracy.
• LiTS provides a valuable, ongoing resource for advancing medical image analysis in liver oncology.