Deep Learning to Classify AL Versus ATTR Cardiac Amyloidosis MR Images

Cardiac amyloidosis comes in two main forms (AL and ATTR) that require different treatments, and distinguishing them on imaging is challenging. This study tested whether a deep-learning algorithm could do it from MRI.

A convolutional neural network analyzing cine MRI images outperformed both a gadolinium-based network and experienced human readers, reaching an AUC of 0.84 versus 0.71 for the best human. The result hints at AI’s potential to extract diagnostic signal that eludes the eye — though the authors are candid that 0.84 accuracy remains short of what clinical deployment would require.

This is a small, retrospective, single-center AI study, and cardiac amyloidosis sits at the periphery of this library’s coronary-disease focus. Its interest is methodological: a proof-of-concept that machine learning may aid a difficult imaging classification.

We rate the evidence moderate-to-limited. It is a competent pilot, but small, narrow, and not yet clinically ready — promising direction rather than practice-changing evidence.