Pneumonia Classification: Assessing the Impact of Fine-Tuning on Diagnostic Accuracy

Pneumonia is a major health problem, particularly in low and middle-income nations. Pneumonia continues to be a serious health concern, especially in areas with limited resources and diagnostic capabilities. With an emphasis on the efficacy of fine-tuning ResNet101, this study assesses the diagnostic performance of deep learning models for pneumonia classification using chest X-ray pictures. The RSNA public dataset was used to train the model, and a locally selected dataset from Bangladesh was used for testing. The adjusted ResNet101 had an overall accuracy of 84% and 89%, with an AUC of 0.86 and 0.67 on RSNA and local data respectively. In contrast, DenseNet121 shown problems with generalization as it declined to 77.6% accuracy (AUC = 0.59) on local data after achieving 95% accuracy (AUC = 0.99) on RSNA. This study also addresses the possibility of deploying diagnostic models via the web and on mobile devices in settings with limited resources. The study shows the viability and constraints of fine-tuned deep learning models for practical implementation in low-resource medical settings, despite the difficulties presented by small, unbalanced local datasets