Experimental Analysis of Radioscopy Images based Pneumonia Disease Detection using Improved Neural Network Methodology

The bacterium Streptococcus pneumonia is the most prevalent causative agent of pneumonia, a severe and potentially fatal lung infection in humans. According to the World Health Organization, pneumonia accounts for one in three deaths in India. Assessing chest radiographs to detect pneumonia typically requires skilled radiologists, making an automatic detection method crucial for rapid treatment, especially in rural areas. This study introduces an Improved Neural Network (INN) learning method, cross-validated against existing Convolutional Neural Networks (CNNs), reflecting the success of deep-learning algorithms in medical image analysis. The research highlights the utility of these methods for disease classification. The features learned by INN and CNN models on large datasets are particularly valuable for image classification tasks. By using feature extractors programmed on trained INN and CNN frameworks, this study compares the performance of various classifiers in identifying normal and abnormal chest radiographs. Statistical results indicate that pre-trained INN models combined with supervised classifier approaches significantly enhance the evaluation of chest radiographs, particularly in detecting pneumonia. This advancement is pivotal for improving diagnosis and treatment outcomes in resource-limited environment.