Left ventricular non-compaction cardiomyopathy automatic diagnosis using a deep learning approach.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Left ventricular non-compaction (LVNC) is a rare cardiomyopathy characterized by excessive trabeculation.
• Diagnosing LVNC often relies on the trabecular volume percentage (VT%) of the left ventricle.

Purpose of the Study:

• To develop and validate a deep learning-based method for measuring VT% and diagnosing LVNC.
• To assess the accuracy and clinical validity of the AI-driven diagnostic approach.

Main Methods:

• A 2D U-Net deep learning model was trained on short-axis cardiac MRI scans from 277 patients.
• The model segmented the left ventricle's cavity, external wall, and trabecular tissue.
• VT% was computed from segmentations, and a threshold was used for diagnosis; 5-fold cross-validation and cardiologist review were performed.

Main Results:

• The U-Net model achieved high segmentation accuracy (Dice coefficients: 0.96 for cavity, 0.89 for wall, 0.84 for trabeculae).
• Cardiologists rated 99.5% of segmentations as clinically valid.
• The AI diagnosis achieved an Area Under the ROC Curve of 0.94, with 87% accuracy, 93% sensitivity, and 80% specificity.

Conclusions:

• Deep learning with U-Net effectively delineates cardiac structures on MRI for accurate LVNC diagnosis.
• This AI solution enables objective, faster analysis, reducing errors and cardiologist workload.
• The method facilitates definitive diagnosis of LVNC through precise hyper-trabeculation measurement.