What is the clinical evidence from this study?

Study design: Other. Population: Cardiovascular disease (n=928). Intervention: Modified version of LeNet-5 model vs. Standard LeNet-5 model. Primary outcome: Classification accuracy.

What does this research mean for the field?

Modifying the LeNet-5 deep learning model by increasing the number of filters improves heart disease prediction accuracy from ECG images by 9.14 percentage points compared to the standard model, achieving 98.38% accuracy. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

December 8, 2022Open Access

Heart Disease Prediction Using Modified Version of LeNet-5 Model

Key Result

The modified version of the LeNet-5 model achieved a classification accuracy of 98.38% for heart disease prediction from ECG images, improving accuracy by 9.14 percentage points compared to the standard LeNet-5 model.

Structured PICO

Does a modified LeNet-5 deep learning model improve the accuracy of heart disease prediction from ECG images compared to the standard LeNet-5 model?

Population

928 ECG images from cardiac patients and normal individuals (normal, abnormal heartbeat, history of myocardial infarction, and infarction patients) sourced from Ch. Pervaiz Elahi Institute of Cardiology Multan, Pakistan.

Intervention

Modified LeNet-5 convolutional neural network model (incorporating 350x350 input size, increased number of filters, maxpooling2D, and dropout of 0.2)

Comparator

Standard LeNet-5 convolutional neural network model (32x32 input size, average pooling, no dropout)

Outcome

Accuracy of heart disease classification from ECG images

A modified LeNet-5 convolutional neural network architecture significantly improves the accuracy of classifying ECG images for heart disease detection compared to the standard model.

Main Result

Absolute Event Rate: 98.38% vs 89.24%

Limitations

The dataset contains a relatively small number of ECG images (928).
The proposed model has not yet been compared with other advanced CNN architectures such as ResNet-50 and DenseNet.

Abstract

Particularly compared to other diseases, heart disease (HD) claims the lives of the greatest number of people worldwide. Many priceless lives can be saved with the help of early and effective disease identification. Medical tests, an electrocardiogram (ECG) signal, heart sounds, computed tomography (CT) images, etc. can all be used to identify HD. Of all sorts, HD signal recognition from ECG signals is crucial. The ECG samples from the participants were taken into consideration as the necessary inputs for the HD detection model in this study. Many researchers analyzed the risk factors of heart disease and used machine learning or deep learning techniques for the early detection of heart patients. In this paper, we propose a modified version of the LeNet-5 model to be used as a transfer model for cardiovascular disease patients. The modified version is compared to the standard version using four evaluation metrics: accuracy, precision, recall, and F1-score. The achieved results indicated that when the LeNet-5 model was modified by increasing the number of used filters, this increased the model's ability to handle the ECGs dataset and extract the most important features from it. The results also showed that the modified version of the LeNet-5 model based on the ECGs image dataset improved accuracy by 9.14 percentage points compared to the standard LeNet-5 model.

Mark Helpful

Bookmark

Relay

View Full Paper