Enhancing pneumonia detection with masked neural networks: a deep learning approach

Pneumonia, a prevalent respiratory disease, affects millions globally. Accurate diagnosis and early detection are essential for managing and treating pneumonia. In recent years, machine learning and visual analysis technologies have shown promise for detecting pneumonia from therapeutic imageries such as chest X-rays. The dataset is collected from a Kaggle and contains X-ray scans of lungs from people of all ages. This dataset includes 5,856 labelled images, of which 4,273 are positive for pneumonia and 1,583 are negative. The data set is preprocessed using data augmentation techniques such as rotation, shifting, shearing, flipping and fill mode. The preprocessed data is trained using a masked neural network (MNN). The essential features are extracted from the last layer of MNN, and then the K-nearest neighbor (KNN) classify the chest X-rays to detect Pneumonia. This study developed a mask generation technique, dropout regularisation, and classifiers to train a model with 98.07% accuracy and minimal losses. This approach could lead to faster and more accurate pneumonia diagnoses, ultimately improving patient outcomes. Our research shows that transfer learning of KNN with MNN can effectively analyse chest X-rays to detect pneumonia.