Enhanced Multi-Class Classification of Lung Diseases in Chest Radiographs using ResNet-50 Based Multi-Perceptive Convolutional Neural Network

Accurate detection of lung diseases is crucial for delivering effective treatment and achieving better patient outcomes. However, traditional diagnostic methods often depend on subjective image interpretation, which can result in errors. This study proposes a multi-class classification approach for lung disorders, including viral pneumonia, COVID-19, and lung opacity, using a ResNet-50 Multi-Perceptive Convolutional Neural Network (ResNet-50-MPCNN) on chest X-rays. The methodology incorporates Contrast Limited Adaptive Histogram Equalization (CLAHE) to expand the image contrast and highlight essential features more effectively. Advanced preprocessing techniques, including normalization, rescaling, and data augmentation, standardize image data and improve training dataset diversity, enhancing the model's robustness. The method was tested using the COVID-19 Radiography Database and showed exceptional accuracy in identifying lung diseases. These results indicate that this approach can support healthcare providers in making more precise and timely diagnoses, leading to enhanced patient care and outcomes.