Deep learning algorithm to improve hypertrophic cardiomyopathy mutation prediction using cardiac cine images

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Hypertrophic cardiomyopathy (HCM) presents diverse genetic phenotypes, necessitating accurate risk stratification.
• Existing risk-stratification systems rely on clinical and echocardiographic data, with limitations in predicting genetic mutations.
• Cardiovascular magnetic resonance (CMR) offers detailed morphological insights valuable for HCM assessment.

Purpose of the Study:

• To enhance mutation-risk prediction in HCM by extracting morphological features using a deep learning (DL) algorithm.
• To evaluate the performance of a DL model in classifying HCM genotypes based on CMR cine images.
• To compare the DL model's predictive accuracy against established genotype scores.

Main Methods:

• Recruited 198 HCM patients, divided into training (147) and test (51) sets.
• Developed a DL model to classify HCM genotypes using nonenhanced four-chamber cine CMR images.
• Assessed established genotype scores (Mayo Clinic I, Mayo Clinic II, Toronto score) and the DL model's performance using Area Under the Curve (AUC).

Main Results:

• The DL model achieved an AUC of 0.80, outperforming Mayo Clinic scores (AUC 0.64-0.70) and the Toronto score (AUC 0.74).
• The DL model demonstrated higher sensitivity (85.71%) and specificity (69.57%) compared to established scores.
• Combining the DL model with the Toronto score yielded the highest predictive performance (AUC = 0.84).

Conclusions:

• Deep learning effectively extracts image features from cine CMR images for HCM genotype classification.
• The DL model based on cine images shows superior performance over established scores in identifying HCM patients with positive genotypes.
• Integrating the DL model with the Toronto score significantly improves the identification of HCM patients with positive genetic mutations.