Abstract Body (Do not enter title and authors here): Background
Maximum wall thickness (MWT) reflects left ventricular hypertrophy, which is associated with adverse prognosis in multiple diseases and at population level. In hypertrophic cardiomyopathy (HCM), MWT is a key marker in risk stratification and is commonly measured by expert clinician annotation. Artificial intelligence (AI) measurement of MWT has shown significantly higher precision than clinicians due to lower test-retest variability, but how this translates to clinical outcome is less understood.

Hypothesis
We hypothesized that a deep-learning AI measurement of MWT obtained from cardiac MRI outperforms expert clinician annotation for the prediction of all-cause mortality.

Methods
We studied a retrospective cohort of 10,382 consecutive patients who underwent cardiac MRI (CMR) at a large referral centre. Mortality data was obtained from a national database. The median follow-up was 5.5 years (IQR 4.4–6.9), during which 1,462 patients (14%) died. A subgroup analysis was conducted in n=1,699 subjects with HCM, of which 207 (12%) died.

AI measurement was performed on all CMR studies, and Cox proportional hazards models were fitted to compare the prognostic value of AI-derived MWT with those derived from clinician annotation used in clinical care.

Results
Compared to clinician annotation, AI-derived MWT had higher prediction for all-cause mortality (p<0.001). Differences remained after adjusting for age and sex. In HCM, MWT was poorly predictive of all-cause mortality (both AI and clinicians C-Indices <0.6) and in these there was no significant difference in mortality prediction (p=0.15). AI analysis took <20s per subject.

Conclusion
Compared to clinician experts, AI-derived MWT shows superior prediction of mortality in all-comer patients. In patients with HCM, AI performance was equivalent. AI measurement is delivered at high speed and low-cost. Findings support the superseding of clinician-derived MWT with AI.