A Neural Network Architecture Using Separable Neural Networks for the Identification of “Pneumonia” in Digital Chest Radiographs

2. Background

Most of the classification in which pneumonia is categorized is based on the country average low income and high income (Agweyu & Lilford 2018) which was presented by the World Health Organization. There are a few guidelines that classify the clinical signs of pneumonia which describe the threshold of risk for children. In this work, we aim to work on the chest radiology images which give a complete report of the threshold data as well as pneumonia classified images. We’ve identified that almost 832 out of 16,031 children die by pneumonia which is 5% of the whole. When it comes to teenagers, out of 11,788, 321 die due to pneumonia, which when summed up, boils down to 3% (Causes of Pneumoniua n.d.). Most of the diagnosis that is carried out in the intensive care configuration is highly unreliable; this includes the diagnosis of the chest radiology manually. There are a few advanced techniques in neural networks dedicated to finding several abnormalities in chest X-Rays. Few include ChexNet (Rajpurkar et al., 2017), a 121-layer neural network architecture that is used to find about 14 abnormalities that also include screening, diagnosis, critical segmentation, and pneumonia classification, and achieved classification accuracy of 0.7680 on the pathology pneumonia. The one main disadvantage is that they take a lot of time to be implemented in rural areas and less populated countries due to minimal availability of the required data. The current practices in detecting pneumonia include manual identification by the radiologist using the weights of the deep neural networks (less precise) for classification, and by using the parameters with respective to the fields of microbiology, etiology, radiology, etc, calculating the entropy to measure the difference between normal respiratory system that’s all fine with the one affected by pathology, and comprehending the snoring patterns to detect for irregularities that might be the causes for pneumonia.