A deep learning-driven pipeline for differentiating hypertrophic cardiomyopathy from cardiac amyloidosis using 2D multi-view echocardiography

Q: What is the clinical evidence from this study?

Study design: Observational. Population: Hypertrophic Cardiomyopathy (HCM) and Cardiac Amyloidosis (CA) (n=1246). Intervention: Multi-view deep learning framework (ResNet-based) vs. Single-view models and traditional classifiers. Primary outcome: Diagnostic performance (precision, sensitivity, specificity, and micro-F1 score) in differentiating HCM, CA, and normal cardiac function.

Q: What does this research mean for the field?

A multi-view deep learning framework applied to 2D echocardiography accurately differentiates hypertrophic cardiomyopathy from cardiac amyloidosis with a sensitivity of 0.81 and specificity of 0.89. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Xiaofeng Li; Bo Peng; Xinyu Li; Z Wang; Hui Deng; Xiaoxian Luo; Lixue Yin; Hongmei Zhang

doi:10.1038/s41598-025-15443-5

October 5, 2026Scientific ReportsOpen Access

A deep learning-driven pipeline for differentiating hypertrophic cardiomyopathy from cardiac amyloidosis using 2D multi-view echocardiography

View Full Paper

Ask AI

Bookmark

Share

Key result

Deep learning on multi-view echocardiography differentiates HCM from cardiac amyloidosis with ~89% specificity.

n=1,246

Why the study?

Hypertrophic cardiomyopathy and cardiac amyloidosis present diagnostic challenges due to overlapping echocardiographic appearances requiring improved differentiation methods.

Does a multi-view deep learning framework applied to 2D echocardiography accurately differentiate hypertrophic cardiomyopathy from cardiac amyloidosis?

Key Points

To develop and evaluate a multi-view deep learning framework that differentiates hypertrophic cardiomyopathy from cardiac amyloidosis using 2D echocardiographic images.
Evaluated an enrolled cohort of 212 patients with hypertrophic cardiomyopathy (HCM), 119 with cardiac amyloidosis (CA), and 200 normal controls between 2018 and 2022.
Pre-processed images with a U-Net model and classified five 2D echocardiographic views: apical 4-chamber, parasternal long axis, and parasternal short axis at the mitral valve, papillary muscle, and apex levels.
Extracted view features with ResNet18 and fused them into a linear network classifier using fivefold cross-validation.
The five-view classification framework achieved a precision of 0.83, sensitivity of 0.81, specificity of 0.89, and a micro-F1 score of 0.82, outperforming single-view models (micro-F1 score of 0.66).
Per-category evaluations yielded precisions of 0.79 for HCM, 0.86 for CA, and 0.84 for normal controls, alongside F1 scores of 0.82, 0.77, and 0.86, respectively.
The image pre-processing U-Net achieved a Dice score of 0.98, and the view classification model reached an overall accuracy of 0.95 across 1,624 ultrasound test images.

Study Design

Type

Observational (n=1,246)

Multicenter

No

Structured PICO

Does a multi-view deep learning framework applied to 2D echocardiography accurately differentiate hypertrophic cardiomyopathy from cardiac amyloidosis?

P

Population

1,246 patients were retrospectively evaluated, from which a cohort of 531 subjects (212 with HCM, 119 with CA, and 200 normal controls) with complete 5-view echocardiograms were selected for the diagnostic disease classification model.

I

Intervention

An automated multi-view deep learning framework for 2D echocardiography (utilizing U-Net for image segmentation, Vision Transformer [ViT] for classifying 5 standard views, ResNet18 for multi-view feature extraction and fusion, and a linear classifier for disease classification).

C

Comparator

Single-view echocardiography models, intermediate multi-view models (2 to 4 views), and alternative machine learning classification architectures (ResNet34, ResNet50, DenseNet121, EfficientNetB2, SVM, logistic regression, random forest, and XGBoost).

O

Outcome

Diagnostic classification performance for differentiating HCM, CA, and normal cardiac function, measured by precision, sensitivity, specificity, and micro-F1 score using fivefold cross-validation.

A multi-view deep learning pipeline can accurately differentiate hypertrophic cardiomyopathy from cardiac amyloidosis using standard 2D echocardiography, providing a potential automated tool to overcome diagnostic challenges caused by overlapping imaging features.

Limitations

Small scale of the dataset used for preliminary validation, potentially limiting generalization.
Lack of multi-center data integration to account for variability in imaging protocols and devices.

Implication

May support automated differentiation of HCM from amyloidosis on routine echocardiography; hypothesis-generating and requires prospective validation before clinical use.

Abstract

Hypertrophic Cardiomyopathy (HCM) and Cardiac Amyloidosis (CA) are two cardiac conditions that can advance to heart failure if left untreated, which present considerable diagnostic challenges due to their overlapping echocardiographic appearance. To address these challenges, this study develops a multi-view deep learning framework that classifies 2D echocardiographic data into five clinically relevant views after image pre-processing: apical 4-chamber, parasternal long axis of left ventricle, parasternal short axis at levels of the mitral valve, papillary muscle, and apex. The framework independently extracts distinctive features from each view, which are then fused for accurate disease classification. The cohort for this study included 212 patients with HCM, 119 with CA, and 200 control subjects with normal cardiac function, enrolled from 2018 to 2022. Utilizing fivefold cross-validation, the model demonstrated precision of 0.83, sensitivity of 0.81, specificity of 0.89, and a micro-F1 score of 0.82. These results affirm the effectiveness of the framework as a reliable diagnostic tool for differentiating between HCM and CA Using a clinical dataset.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Li et al. (2026) conducted an observational in Hypertrophic Cardiomyopathy (HCM) and Cardiac Amyloidosis (CA) (n=1,246). Multi-view deep learning framework (ResNet-based) vs. Single-view models and traditional classifiers was evaluated on Diagnostic performance (precision, sensitivity, specificity, and micro-F1 score) in differentiating HCM, CA, and normal cardiac function. A deep learning framework analyzing five 2D echocardiographic views differentiated hypertrophic cardiomyopathy from cardiac amyloidosis with a precision of 0.83, sensitivity of 0.81, specificity of 0.89, and a micro-F1 score of 0.82.

synapsesocial.com/papers/6ac5e0deb2dceb1c429fb544 https://doi.org/10.1038/s41598-025-15443-5

Topics

Cardiac amyloidosis

Artificial intelligence in cardiology

View Full Paper

Ask AI

Bookmark

Share

A deep learning-driven pipeline for differentiating hypertrophic cardiomyopathy from cardiac amyloidosis using 2D multi-view echocardiography

Key Points

Study Design

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Topics

A deep learning-driven pipeline for differentiating hypertrophic cardiomyopathy from cardiac amyloidosis using 2D multi-view echocardiography

Key Points

Study Design

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

Topics

Keep reading