Artificial intelligence-based cardiac transthyretin amyloidosis detection and scoring in scintigraphy imaging: multi-tracer, multi-scanner, and multi-center development and evaluation study.

Summary

A fully automated deep-learning pipeline trained and tested across six datasets (12 cameras) detected and scored ATTR-CM on whole-body scintigraphy with high performance, achieving external AUCs up to 1.00 for detection and up to 0.96 for scoring. Explainability maps focused on clinically relevant cardiac areas, supporting model validity and potential for earlier diagnosis.

Key Findings

• Internal test performance exceeded AUC 0.95 and F1 0.90 for both detection and scoring.
• External validation achieved detection AUCs of 0.93, 0.95, and 1.00; scoring AUCs of 0.95, 0.83, and 0.96.
• Explainability (Grad-CAM/saliency) highlighted clinically relevant cardiac regions; prospective flagging identified additional possible cases.

Clinical Implications

AI-assisted screening of total-body scintigraphy could flag probable ATTR-CM for confirmatory evaluation, standardize Perugini-like scoring across centers, and reduce diagnostic delays.

Limitations

• Retrospective datasets with incomplete reference standards; limited biopsy confirmation
• Generalizability to unseen tracers/cameras and prospective workflow integration requires testing

Future Directions

Prospective, multi-center clinical utility studies; harmonization across acquisition protocols; head-to-head comparisons with expert readers and hybrid AI-human workflows.