A Deep Learning-Driven Pipeline for Differentiating Hypertrophic Cardiomyopathy from Cardiac Amyloidosis Using 2D Multi-View Echocardiography

TL;DR

The paper tackles the diagnostic challenge of distinguishing two cardiomyopathies, HCM and CA, which often present with similar echocardiographic features. It introduces a three-stage deep learning pipeline that first classifies five 2D echocardiography views with a Vision Transformer, then extracts per-view features using a modified ResNet18, fusing them for final classification among HCM, CA, and Normal. The approach achieves a micro-F1 score of 0.904 and precision/recall of 0.905 in differentiating HCM and CA on a private dataset, with view classification accuracy of 0.95, demonstrating the benefits of multi-view fusion and interpretability via Grad-CAM. This method offers a scalable, automated tool to aid clinicians in rapid, objective differentiation of these conditions in routine echocardiography.

Abstract

Hypertrophic cardiomyopathy (HCM) and cardiac amyloidosis (CA) are both heart conditions that can progress to heart failure if untreated. They exhibit similar echocardiographic characteristics, often leading to diagnostic challenges. This paper introduces a novel multi-view deep learning approach that utilizes 2D echocardiography for differentiating between HCM and CA. The method begins by classifying 2D echocardiography data into five distinct echocardiographic views: apical 4-chamber, parasternal long axis of left ventricle, parasternal short axis at levels of the mitral valve, papillary muscle, and apex. It then extracts features of each view separately and combines five features for disease classification. A total of 212 patients diagnosed with HCM, and 30 patients diagnosed with CA, along with 200 individuals with normal cardiac function(Normal), were enrolled in this study from 2018 to 2022. This approach achieved a precision, recall of 0.905, and micro-F1 score of 0.904, demonstrating its effectiveness in accurately identifying HCM and CA using a multi-view analysis.

Figures (9)

• Figure 1: The schematic diagram for criteria of inclusion and exclusion criteria patients. The number of patients used for each step is indicated. ①②③ is the criterion. ① Left ventricular wall thickness ≥15mm in any segment or ≥12mm in patients with family history. ② Suspected CA by echocardiography, confirmed amyloidosis by tissue biopsy, or confirmed by late gadolinium enhanced cardiac magnetic resonance (CMR) imaging. ③ With all 5 views.
• Figure 2: Cardiac imaging of five views. The second row is the short axis of arasternal left ventricular. a) A4C: apical 4-chamber, b) PLAX: Long axis of parasternal left ventricula, c) PSAX_MV: short axis at mitral valve, d) PSAX_MP: short axis at papillary muscle, e) PSAX_AC: short axis at apical
• Figure 3: The pipeline for fully automated disease classification. The view classification model (A), the feature extraction model (B), three type disease classifier (C) and the network structure of feature extraction (D).
• Figure 4: Transformer architecture for echocardiogram view classification. The figure illustrates how each echocardiogram is divided into several patches. These patches are then converted into linear embeddings, effectively simulating the sequence-based input that is a standard approach in NLP.
• Figure 5: Model of View classification successfully discriminate echocardiographic views. A. T-SNE visualization of view classification depicts the successful grouping of test images corresponding to 5 different echocardiographic views. B. Confusion matrix demonstrating successful and unsuccessful probability of view classifications within the test data set. Numbers along the diagonal represent successful probability of classifications, whereas off-diagonal entries are misclassifications.