Conference Coverage

AI Echocardiography Model Shows High Negative Predictive Value for Detecting ATTR-CM in Diverse Heart Failure Cohort

Edited by:
Kate Young

Key Highlights

  • An AI model analyzing apical 4-chamber echocardiography clips demonstrated an area under the curve of 87.7% for detecting transthyretin cardiac amyloidosis.
  • The model achieved a high negative predictive value of 97.8% in a prospective cohort of Black and Hispanic patients with heart failure.
  • Findings suggest the tool may help rule out ATTR-CM and identify patients who would benefit from confirmatory scintigraphy testing.

New prospective data presented at the American College of Cardiology’s Annual Scientific Session in New Orleans, LA evaluated whether an echocardiography-based artificial intelligence model could improve identification of transthyretin cardiac amyloidosis (ATTR-CM) in clinical practice. The findings come from the SCAN-MP study and assessed the performance of a previously developed deep learning algorithm in a prospective screening population.

The investigators evaluated a fully automated deep learning model designed to detect cardiac amyloidosis from a single apical 4-chamber transthoracic echocardiography video. The algorithm uses a 3-dimensional convolutional neural network and had previously demonstrated high diagnostic accuracy in retrospective datasets with high disease prevalence. In the current analysis, researchers sought to determine whether the model maintained performance in a lower-prevalence prospective cohort undergoing screening for ATTR-CM.

The study included 534 Black and Hispanic patients aged older than 60 years with heart failure enrolled in the SCAN-MP cohort. All participants underwent baseline transthoracic echocardiography, technetium-pyrophosphate (Tc-PyP) scintigraphy imaging, and transthyretin (TTR) genotyping. Model predictions were generated from echocardiographic apical 4-chamber video clips and categorized using a Youden index threshold of 0.05.

Study Findings

Among the 534 participants analyzed, 43 patients (8%) were diagnosed with ATTR-CM based on Tc-PyP imaging. The AI model demonstrated a sensitivity of 76.7% and a specificity of 89.2%. Positive predictive value was 38.4%, while the negative predictive value reached 97.8%. Overall diagnostic discrimination yielded an area under the curve (AUC) of 87.7%.

Researchers also compared echocardiographic parameters between patients correctly identified by the model and those missed by the algorithm. There were no significant differences in ejection fraction, ventricular wall thickness, or global longitudinal strain between patients classified as true positives (n=33) and those classified as false negatives (n=10).

Clinical Implications

According to the study authors, the high negative predictive value observed in this prospective cohort supports the potential role of the AI model as a screening tool to rule out ATTR-CM in patients with heart failure. The researchers noted that identifying patients with a high predicted probability of disease could help guide additional diagnostic evaluation using nuclear scintigraphy imaging, where testing would have a higher diagnostic yield.

Expert Commentary

“In a low-prevalence cohort, the AI model showed high negative predictive value, supporting its role in ruling out ATTR-CM and identifying a high-risk cohort in whom further testing with scintigraphy would have a high yield,” the researchers concluded.

Reference
Ryan G, Teruya SL, Maurer MS, et al. Prospective evaluation of an echocardiography-based artificial intelligence model for detecting transthyretin cardiac amyloidosis: results from the SCAN-MP study. Presented at: American College of Cardiology Annual Scientific Session (ACC.26); March 28, 2026; New Orleans, LA. https://cattendee.abstractsonline.com/meeting/21230/presentation/4687