CHESTX-PNEUMONET: OPTIMIZED CNN WITH NOVEL LOSS FUNCTION FOR PNEUMONIA DETECTION

The automated and accurate detection of pneumonia poses a significant problem in medical analysis due to the subtlety of its indicators in X-ray or CT images. This task is of paramount importance, as pneumonia claims millions of lives annually. Leveraging sophisticated techniques such as deep learning is essential to enhancing both diagnostic precision and operational efficiency. However, adapting existing neural network frameworks for clinical imaging tasks often results in overfitting and limited transferability. To mitigate these limitations, we introduce PneumoClassifyNet, an innovative lightweight convolutional network architecture tailored specifically for Pneumonia Detection in chest X-ray analysis. This architecture is more compact yet more potent than traditional fine-tuning approaches. Additionally, we introduce a new loss function, RadCE-loss, designed to effectively extract distinguishing characteristics from incorrectly classified and fuzzy images. Furthermore, the convolutional kernels are optimized within the convolutional neural network (CNN) model to improve accuracy of classification. The paper presents PneumoClassifyNet, an optimized CNN for chest X-ray classification, and RadCE-loss, a function that enhances accuracy by handling misclassified images. Results of Experiments indicate that the lightweight PneumoClassifyNet, coupled with the RadCE-loss, achieves superior performance across key metrics, including F1-score, recall, accuracy and AUC. These findings affirm that a carefully optimized convolutional neural network (CNN) architecture can outperform fine-tuned deep learning models.