Liver Segment and Lesion Segmentation on CT and MRI: An Open-Source Contribution to TotalSegmentator.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate segmentation of liver anatomy and lesions is crucial for diagnosis and treatment monitoring in liver diseases.
• Manual segmentation is time-consuming and subject to inter-observer variability.
• Deep learning offers potential for automated and accurate medical image analysis.

Purpose of the Study:

• To develop and validate a deep learning-based tool for automatic liver and liver lesion segmentation on CT and MRI.
• To assess the clinical utility of the tool in evaluating hepatocellular carcinoma (HCC) response to transarterial chemoembolization (TACE).
• To provide the developed models as open-source software to enhance existing tools like TotalSegmentator.

Main Methods:

• Liver segmentation models were trained and tested using fivefold cross-validation on 193 CTs and 120 MRIs.
• Liver lesion segmentation models (nnU-Net) were trained on 750 images and tested on 80, using a dataset of 414 CTs and 308 MRIs.
• Inter-rater variability was assessed on 20 CTs and 20 MRIs; clinical utility was evaluated on 172 TACE-treated HCC cases.

Main Results:

• The liver segmentation model achieved Dice coefficients of 0.897 for CT and 0.847 for MRI.
• Liver lesion detection on CT showed 75.8% sensitivity and 0.658 Dice; on MRI, 62.7% sensitivity and 0.337 Dice.
• The model demonstrated potential in tracking HCC treatment response, with a significant decrease in median HCC attenuation post-TACE.

Conclusions:

• The developed deep learning algorithms reliably detect and segment liver segments and lesions on both CT and MRI.
• The tool shows potential clinical value for research and applications, particularly in monitoring treatment response for HCC.
• The open-source release of these models aims to improve capabilities in liver image analysis and segmentation.