Overview

Cardiac amyloidosis is a progressive cardiomyopathy with poor prognosis, and the emergence of disease-modifying therapies has made early diagnosis increasingly important. Because definitive diagnosis requires multiple confirmatory tests, an accessible and accurate screening approach is essential. Echocardiography plays a pivotal role in screening, and AI-based fully automated echocardiographic analysis has been developed to detect cardiac amyloidosis from a single apical four-chamber cine clip. This study compared the screening performance of the Us2.ai deep learning model against an automated increased wall thickness score derived from Us2.ai measurements, using endomyocardial biopsy as the reference standard.

 

Study Design

This retrospective study screened 171 consecutive patients who underwent endomyocardial biopsy at Tokushima University Hospital between 2016 and 2025. Patients with AI-measured LV wall thickness of 12 mm or above were included in the analysis, as this threshold is required for both the AI-CNN and AI-IWTS approaches. Of the 92 patients meeting this threshold, 58 had biopsy-confirmed cardiac amyloidosis including 48 ATTR, 9 AL, and 1 ApoA-1. Diagnostic performance was evaluated using sensitivity, specificity, predictive values, and c-index with 95% confidence intervals.

 

Key Results

AI Deep Learning Model (AI-CNN):

  • C-index 0.82 (95% CI 0.72 to 0.89)
  • Sensitivity 98%, specificity 65%
  • PPV 83%, NPV 96%

Automated Wall Thickness Score (AI-IWTS):

  • C-index 0.59 (95% CI 0.48 to 0.69)
  • Sensitivity 26%, specificity 91%
  • PPV 83%, NPV 43%

McNemar's test confirmed a significant difference in classifications between the two methods (p<0.001), with the AI-CNN demonstrating substantially higher sensitivity and NPV.

 

Why This Matters

Screening for cardiac amyloidosis requires high sensitivity to avoid missing patients who may benefit from early treatment. This study demonstrates that the Us2.ai deep learning model, operating from a single apical four-chamber clip, achieves near-perfect sensitivity at 98% in a biopsy-proven cohort, with an NPV of 96%, making it a strong tool for ruling out cardiac amyloidosis in patients with wall thickening. The substantially superior performance over an automated scoring approach based on wall thickness measurements underlines the value of end-to-end deep learning over measurement-derived indices for complex pattern recognition.

 

Conclusion

In a biopsy-referenced cohort, the Us2.ai AI-CNN model showed substantially higher sensitivity and discrimination than the automated IWT score, suggesting it may serve as a practical echocardiographic screening tool to support early detection of cardiac amyloidosis in routine clinical practice.

 

 

Nomura, Y., Hirata, Y., Nishio, S., Saijo, Y., Nonaka, R., Yamaguchi, N., Zheng, R., Okushi, Y., Takahashi, T., Kusunose, K., Yamada, H., Sata, M. (2026, August 28-31). Screening performance of a deep learning model versus an automated wall thickness score in biopsy-proven cardiac amyloidosis [Poster presentation abstract]. ESC Congress 2026, Munich, Germany.