Decoding the Heart Through Computed Tomography: Early Cardiomyopathy Detection Using Ensemble-Based Segmentation and...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Cardiomyopathy diagnosis often relies on late-stage symptoms, delaying treatment.
• There is a need for early, non-invasive screening methods for subclinical myocardial disease.
• Clinically indicated CT scans offer an opportunity for opportunistic screening.

Purpose of the Study:

• To develop an automated framework using CT scans for left ventricular myocardium segmentation.
• To estimate the probability of underlying myocardial disease using radiomic feature analysis.
• To create an Ensemble Machine Learning (ML) model for early cardiomyopathy detection.

Main Methods:

• Trained ML models on 60 CT scans for left ventricular myocardium segmentation.
• Developed a novel Ensemble model combining four top-performing segmentation models (Unet++, ED w/ASC, FPN, TresUNET).
• Performed radiomic feature analysis on manually and automatically segmented CT scans from 100 unseen cases.

Main Results:

• The Ensemble model achieved a DICE score of 0.882 post-optimization and 0.907 on external validation.
• Identified 15 key radiomic predictors of myocardial disease.
• The model showed strong performance in detecting myocardial disease with AUCs of 0.85 (manual) and 0.8 (automatic).

Conclusions:

• Presents a fully automated CT-based framework for myocardial segmentation and radiomic phenotyping.
• Accurately estimates the probability of underlying myocardial disease, demonstrating generalizability.
• Highlights AI-driven CT analysis potential for early, non-invasive population-level cardiomyopathy screening.