Automatic MRI segmentation of pectoralis major muscle using deep learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate measurement of Pectoralis Major Muscle Cross-Sectional Area (PMM-CSA) is crucial for clinical assessments.
• Manual segmentation and measurement of PMM-CSA from MRI are time-consuming and prone to variability.

Purpose of the Study:

• To develop and validate a deep convolutional neural network (CNN) for automated Pectoralis Major Muscle (PMM) segmentation.
• To create a CNN-based method for selecting the greatest PMM Cross-Sectional Area (PMM-CSA) from axial MRI slices.
• To compare the CNN method's performance against manual reference standards.

Main Methods:

• A deep CNN (MONAI/Pytorch SegResNet) was trained for PMM segmentation on 134 axial T1-weighted MRIs.
• The segmentation model achieved a Mean Dice score of 0.94 ± 0.01 on internal test cases.
• OpenCV2 was used to calculate PMM-CSA from model predictions, with top-3 slice accuracy evaluated against ground truth.

Main Results:

• The CNN segmentation model demonstrated high accuracy with a Mean Dice score of 0.94 ± 0.01.
• The PMM-CSA selection method achieved a top-3 accuracy greater than 98% on internal test cases.
• The combined CNN approach accurately automated PMM segmentation and PMM-CSA selection.

Conclusions:

• A deep CNN method can accurately segment PMM and select the greatest PMM-CSA on axial MRI.
• This automated approach offers a precise and efficient alternative to manual measurements.
• The validated CNN method holds potential for improving clinical assessments and research involving PMM.