Detection of sarcopenia using deep learning-based artificial intelligence body part measure system (AIBMS)

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Gerontology

Background:

• Sarcopenia, an aging-related muscle loss syndrome, elevates risks of falls, fractures, disability, and mortality.
• Current sarcopenia diagnosis via medical imaging demands laborious manual segmentation of abdominal body parts.
• Efficient automated methods for body part segmentation and disease prediction are crucial.

Purpose of the Study:

• To develop an Artificial Intelligence Body Part Measure System (AIBMS) for automated body part segmentation and quantification from abdominal CT scans.
• To assess the performance of deep learning models (SEG-NET, U-NET, Attention U-NET) for segmentation accuracy.
• To construct and validate an AI-based model for sarcopenia prediction using automated measurements.

Main Methods:

• Designed an AI system (AIBMS) utilizing deep learning models (SEG-NET, U-NET, Attention U-NET) for abdominal CT scan analysis.
• Trained and evaluated segmentation models on diverse datasets, achieving high accuracy (DSC score > 0.9).
• Developed a sarcopenia classification model (Auto SMI) using optimized cutoff values (threshold 40.69) based on Youden index, achieving an AUC of 0.874.

Main Results:

• The AIBMS achieved high accuracy in segmenting body parts from abdominal CT scans, with DSC scores exceeding 0.9.
• Recommendations for selecting appropriate deep learning models were provided for different clinical scenarios.
• The Auto SMI model demonstrated high accuracy in predicting sarcopenia (AUC = 0.874), with an optimized threshold of 40.69.

Conclusions:

• An AI system was successfully developed for automated body part segmentation in abdominal CT images.
• The developed system enables accurate sarcopenia prediction through an AI-based model utilizing optimized cutoff values.
• This AI approach offers an efficient and accurate solution for sarcopenia diagnosis and management.