Optimisation of deep learning-based models for the diagnosis of heart disease through ODTH method

In middle- and low-income countries, cardiovascular illnesses (CVDs) constitute the leading cause of death, with heart attacks and strokes accounting for around 80% of CVD-related fatalities. Enabling early intervention and treatment planning, effective cardiac irregularity prediction and the design of trustworthy heart disease prediction systems eventually lower death rates. This research investigated the viability of predicting cardiac disease using tabular data and convolutional neural networks (CNN). We first retrieved pertinent data from the collection of records, which was then abridged to 14 characteristics; each record is converted into heatmaps, and PNG files of the heatmaps are stored for further CNN processing and visualisation to DenseNet121, ResNet50 and VGG19. Using 10-fold cross-validation, we discovered that DenseNet121, in addition to the optimisation method stochastic gradient descent (SGD), performed better with 97% accuracy while the other two VGG19 54.39% and ResNet50is 51.00% models, performed low as compared to DenseNet121 in addition with the use of accuracy of 54.39% and 51.00%, respectively. Our research demonstrates that deep learning models are capable to correctly forecast heart disease from tabular data. In this paper, it is concluded that tabular data can be given as input to deep learning models to achieve better accuracy and good results can be observed for further study in the field of disease prediction.