CinE caRdiac magneTic resonAnce to predIct veNTricular arrhYthmia (CERTAINTY)

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Identifying patients at low risk of ventricular arrhythmia (VA) is crucial for implantable cardioverter-defibrillator (ICD) candidates to prevent unnecessary procedures and complications.
• Current risk stratification models for primary prevention ICD implantation have limitations.

Purpose of the Study:

• To develop and validate deep learning models for predicting VA risk using cine cardiac magnetic resonance (CMR) imaging.
• To assess the utility of "cine fingerprint" features extracted from CMR for VA risk prediction.

Main Methods:

• The CERTAINTY study utilized a training cohort of 350 primary prevention ICD recipients.
• Two deep learning neural networks were developed: a Cine Fingerprint Extractor (unsupervised feature extraction) and a Risk Predictor (outputting a cine risk score).
• Models were trained on cine CMR data acquired immediately prior to ICD implantation.

Main Results:

• Patients who experienced VA (n=96) had a significantly higher cine risk score compared to those without VA.
• The cine risk score was independently associated with VA risk after adjusting for clinical factors and CMR-derived scar extent.
• Deep learning analysis of cine CMR identified features predictive of VA.

Conclusions:

• Non-contrast cine CMR contains inherent features that can significantly improve VA risk prediction in primary prevention ICD candidates.
• The developed deep learning approach offers a novel method for enhanced VA risk stratification.
• External validation across multiple centers is solicited to confirm generalizability.