Automated Abdominal Segmentation of CT Scans for Body Composition Analysis Using Deep Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate body composition quantification from CT scans is crucial for clinical assessments.
• Manual segmentation is time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To develop and validate a fully automated algorithm for abdominal segmentation on CT images.
• To assess the algorithm's accuracy in quantifying body composition metrics.

Main Methods:

• A U-Net based convolutional neural network was trained on 2430 CT examinations.
• The algorithm was tested on independent datasets, including patients with hepatocellular carcinoma (HCC).
• Segmentation performance was evaluated using Dice scores and compared to expert manual segmentation.

Main Results:

• The algorithm achieved high Dice scores across subcutaneous, muscle, and visceral adipose tissue compartments (e.g., 0.98 ± 0.03 for subcutaneous fat).
• Performance met or exceeded expert manual segmentation accuracy.
• The model demonstrated good generalization across different abdominal levels and patient cohorts.

Conclusions:

• The developed algorithm provides accurate and automated abdominal segmentation for body composition analysis.
• This automated approach surpasses manual segmentation in performance and efficiency.
• The model is capable of 3D body composition quantification from CT examinations.