Evaluating the Performance of Efficientnet Variants in Pneumonia Detection Using Medical Image Data

Pneumonia remains one of the leading causes of death throughout the world, especially in less developed nations where diagnostic centers of healthcare are simply not available. Traditional methods for diagnosis are timeconsuming and dependent upon skilled radiologists, yet diagnosis at an early stage decreases mortality and saves lives. Deep learning using Convolutional Neural Networks (CNN) has recently shown to be highly accurate in classifying medical images. Comparative performance of EfficientNet models B0B7 in classifying pneumonia using chest X-ray images is assessed in this study. The dataset in this study includes <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{5, 8 6 3}$</tex> images labeled as NORMAL and PNEUMONIA. Optimal hyperparameters transfer learning was implemented to train all the models to perform with high accuracy for classification. Results indicate EfficientNet B1 had 96.17% accuracy, 95.14% F1-score, and 97.41 % recall with highest sensitivity for detecting pneumonia cases. Receiver Operating Characteristic (ROC) curve analysis also demonstrates the models' stability, as all AUCs are greater than 0.98. The findings suggest deep learning models, as in this research, EfficientNet models can provide accurate computer-aided diagnostic support for the detection of pneumonia. Some of the areas of future research involve enhancing model interpretability, expanding the data sets to represent diverse populations, and deploying AI diagnostic tools in clinical environments. Application of such models in resource-poor environments has the potential to make a significant difference in early pneumonia detection, decreasing world disease and death burdens.