Abstract

Artificial intelligence (AI) plays a vital role in both the modern digital world and the medical field. Within healthcare, the diagnostic system is rapidly gaining prominence as it offers clear insights for radiologists and patients alike. However, the automation of medical image processes is challenging for domain experts due to the enormous volume of data generated by traditional systems. Recently, deep learning algorithms have emerged as powerful techniques for investigating medical images, capable of performing various tasks such as identification, classification, prediction, and pattern recognition. Despite ongoing research efforts, medical image classification tasks continue to face various challenges and obstacles that impact the performance and accuracy of classification models. This study focused on the reduce the feature extraction problem in medical image classification by applying fine-tune the hyper parameters. The feature extraction problem requires the development of sophisticated deep models which can capture higher-level abstractions from raw data. In this study, we have taken a significant step by enhancing three pre-trained deep learning models. The proposed methodology focus was on handling a large dataset of multi-class COVID-19 chest X-ray images and explored the three models namely VGG Net, EfficientNet, and InceptionNet. Analysing the outcomes of experiments, each model yielded different results in terms of accuracy and loss rates. The accuracy we achieved for VGG Net was 86%, for EfficientNet it was 93% and for Inception Net was 78% and correspondingly, the loss rates for these models were 0.3, 0.1, and 0.5, respectively. This observation positions the EfficientNet model as a particularly suitable candidate for effectively classifying large-scale medical image datasets with better accuracy and low loss rate.