Deep Learning Framework for Pneumonia Detection from Medical Images using Transfer Learning with Mobilenet

Abstract: Pneumonia remains a leading cause of morbidity and mortality worldwide, particularly in regions with limited access to expert radiological assessment, highlighting the need for reliable automated diagnostic support systems. This study presents a deep learning–based framework for binary classification of chest X-ray images into pneumonia and normal categories using a pretrained convolutional neural network. A dataset comprising 5,856 chest X-ray images was utilized, including 4,273 pneumonia and 1,583 normal cases, with an imbalanced class distribution. The data were partitioned into training, validation, and held-out test sets with approximately 79%, 5%, and 16% of the data, respectively. To mitigate data scarcity and improve generalization, controlled data augmentation was applied during training.A pretrained MobileNet architecture was employed as the feature extractor, leveraging transfer learning to adapt to the medical imaging domain. The model was trained using mini-batch optimization with a batch size of 32 and input resolution of 224 × 224 pixels. Performance evaluation was conducted on an independent test set using multiple metrics, including accuracy, precision, recall, F1-score, and receiver operating characteristic area under the curve.The proposed framework achieved a test accuracy of 86.07%, with a precision of 83.88%, recall of 99.06%, and F1-score of 90.84%. The receiver operating characteristic analysis yielded an area under the curve of 0.96, indicating strong discriminative capability. The confusion matrix analysis revealed a low false negative rate, demonstrating the model’s effectiveness in identifying pneumonia cases, although a relatively higher false positive rate suggests limitations in distinguishing normal cases. Overall, the results indicate that the model is well-suited for high-sensitivity screening scenarios, where minimizing missed pneumonia cases is critical.