Lightweight Transfer Learning Models for Covid-19 Pneumonia Identification Using Chest X-Ray Imaging

Abstract

The increasing rate of spreading COVID-19 created a serious need in accurate and extensive diagnostic centers to support the process of clinical diagnosis, particularly in health institutions that are over-allocated with resources. The chest X-ray is still considered as one of the most readily available procedures to assess the lungs; however, the process of the interpretation is tedious and skewed by inter-observer variations. In this paper, automated deep learning methods will be employed to solve the issue of efficiently identifying pneumonia cases using chest X-ray outcomes. The main objective of it is to compare the various deep learning frameworks with the view of determining the most appropriate one to detect pneumonia. The framework of this paper is a comparison framework with convolutional neural network-based architectures along with a baseline CNN, EfficientNet and Lightweight MobileNet, and Lightweight EfficientNet. The models are trained and tested on a selected chest X-ray dataset with regular preprocessing, data augmentation, and equal spread of classes. The experimental findings prove that lightweight architectures outperform standard CNN models in terms of high accuracy and optimal sensitivity. Lightweight EfficientNet model is the most qualified model by the total performance of 92.5% accuracy, 92.4% precision, 90.1% recall and 90.2% F1-score, which means that the classification performance is strong. The results relate to the effectiveness of lightweight DL models in obtaining differentiating lung characteristics related to COVID-19 pneumonia. The proposed comparative study demonstrates that optimized lightweight deep learning models can provide accurate, rapid, and clinically viable results in detecting COVID-19-related pneumonia from chest X-ray images, thereby validating their potential integration into clinical diagnostic protocols.