A two-step convolutional neural network based computer-aided detection scheme for automatically segmenting adipose...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate adipose tissue volume assessment is crucial for disease risk prediction, diagnosis, and prognosis.
• Current methods relying on single CT slices are subjective and limited.
• Volumetric CT data offers a more comprehensive view of adipose tissue distribution.

Purpose of the Study:

• To develop and validate a deep learning-based computer-aided detection (CAD) scheme for automated segmentation of subcutaneous fat areas (SFA) and visceral fat areas (VFA) from volumetric CT images.
• To overcome the limitations of manual selection of CT slices for fat area estimation.
• To enhance the accuracy and efficiency of adipose tissue quantification in clinical settings.

Main Methods:

• A two-step deep learning framework utilizing two Convolutional Neural Networks (CNNs): Selection-CNN and Segmentation-CNN.
• Selection-CNN was trained on 2,240 CT slices to identify relevant abdominal slices containing SFA and VFA.
• Segmentation-CNN was trained on 84,000 pixel patches to classify fat pixels into SFA and VFA categories within the selected slices.

Main Results:

• The Selection-CNN achieved 95.8% accuracy in selecting appropriate CT slices for analysis.
• The Segmentation-CNN demonstrated 96.8% accuracy in segmenting fat pixels into SFA and VFA.
• High agreement was observed between the automated segmentation results and manual segmentation.

Conclusions:

• The developed deep learning-based CAD scheme is feasible for automated abdominal CT recognition and SFA/VFA segmentation.
• The system achieves high accuracy, comparable to manual segmentation, offering a reliable tool for adipose tissue assessment.
• This technology holds potential for improved disease risk prediction and patient management through precise fat volume quantification.