Pic2Diagnosis: A Method for Diagnosis of Cardiovascular Diseases from the Printed ECG Pictures

TL;DR

The paper introduces a digitization-free method to diagnose cardiovascular diseases directly from printed ECG images using a two-step curriculum learning approach: first train on segmentation masks derived from synthetic data, then fine-tune on grayscale inverted ECG images. An ensemble of three model variants improves robustness and achieves high AUROC and F1 scores on the BHF ECG Challenge and validation sets, outperforming individual models. The approach is designed to handle real-world artifacts common in resource-limited settings, enabling rapid, accessible CVD screening without digitization. Interpretability via XGrad-CAM indicates complementary focus among ensemble members, supporting reliable integration into clinical workflows with potential for further explainability enhancements.

Abstract

The electrocardiogram (ECG) is a vital tool for diagnosing heart diseases. However, many disease patterns are derived from outdated datasets and traditional stepwise algorithms with limited accuracy. This study presents a method for direct cardiovascular disease (CVD) diagnosis from ECG images, eliminating the need for digitization. The proposed approach utilizes a two-step curriculum learning framework, beginning with the pre-training of a classification model on segmentation masks, followed by fine-tuning on grayscale, inverted ECG images. Robustness is further enhanced through an ensemble of three models with averaged outputs, achieving an AUC of 0.9534 and an F1 score of 0.7801 on the BHF ECG Challenge dataset, outperforming individual models. By effectively handling real-world artifacts and simplifying the diagnostic process, this method offers a reliable solution for automated CVD diagnosis, particularly in resource-limited settings where printed or scanned ECG images are commonly used. Such an automated procedure enables rapid and accurate diagnosis, which is critical for timely intervention in CVD cases that often demand urgent care.

Figures (9)

• Figure 1: This figure outlines the proposed framework for multi-label cardiovascular disease diagnosis from ECG images. The pipeline begins with preprocessing steps, including CLAHE for contrast enhancement, background removal, perspective transformation, grayscale conversion, and color inversion. The processed image is passed through three independently trained models, and their logits are combined using an ensemble approach. Finally, binarization with per-class thresholds is applied to produce a 1/0 output for each class, indicating the presence or absence of myocardial infarction (MI), ST-T changes (STTC), conduction disorders (CD), hypertrophy (HYP), and atrial fibrillation (AF).
• Figure 2: An example of a synthetically generated image from the GenECG dataset.
• Figure 3: Simplified convex hull and corner points on the example image.
• Figure 4: The processed example image after applying preprocessing steps, including background removal and perspective transformation.
• Figure 5: An example image generated by the ECG-image-kit using an ECG signal from the PTB-XL dataset.