Using a new artificial intelligence-aided method to assess body composition CT segmentation in colorectal cancer...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Assessing body composition in colorectal cancer (CRC) patients is crucial for treatment planning and prognosis.
• Computed tomography (CT) scans provide valuable data for body composition analysis, particularly at the L3 vertebral level.
• Automated segmentation using artificial intelligence (AI) offers a potential solution for efficient and accurate body composition assessment.

Purpose of the Study:

• To evaluate the accuracy of an AI-generated model for automated segmentation and quantification of body composition from CT slices.
• To assess the performance of the AI model in analyzing muscle, visceral adipose tissue (VAT), and subcutaneous adipose tissue (SAT) at the L3 region in CRC patients.

Main Methods:

• A U-Net convolutional neural network was trained on 338 CT slices from 319 CRC patients.
• Manual segmentation served as the ground truth for training and validation.
• Segmentation performance was assessed using Dice similarity coefficients on validation (68 slices) and test (203 slices) datasets.

Main Results:

• The AI model achieved excellent segmentation performance, with Dice similarity coefficients >0.98 for validation and >0.97 for test datasets.
• Strong positive correlations (Spearman's correlation coefficients: 0.944-0.999) were observed between manual and AI measurements of body composition.
• The AI system demonstrated high accuracy in quantifying cross-sectional area and Hounsfield unit density for muscle, VAT, and SAT.

Conclusions:

• The fully automated AI segmentation system demonstrates high accuracy in assessing and quantifying abdominal muscle and adipose tissues from L3 CT slices in CRC patients.
• This AI model serves as a reliable tool, comparable to the gold standard, for body composition analysis in clinical settings.
• The study validates the utility of AI in streamlining body composition assessment, aiding in the management of colorectal cancer.