International Journal of Computer Science and Technology
IJCST Vol 13 Issue 2 (April-June 2022)

S.No. Research Topic Paper ID Download

Pneumonia Detection in Pediatric Chest X-Rays with Capsule Neural Networks

Robert Langenderfer, Ezzatollah Salari, Jared Oluoch


Pneumonia is a common respiratory infection caused by bacteria, viruses, chemicals, or fungi. In 2019, more than 2.5 million people died from this disease, and is the single largest cause of death in children under the age of five globally. Diagnosis of pneumonia typically employs chest radiography, which is visually interpreted by highly trained radiologists. Given the cost, unavailability, and fallibility of radiologists, there has been significant interest in developing machine learning models to automate the diagnostics process. Recent research has focused on Deep Learning Neural Network (DNN) and Convolutional Neural Network (CNN) models to perform medical diagnostic classification. However, in this study we deployed a capsule based neural network for the detection of pneumonia in pediatric chest X-ray images. Where traditional CNN models discard significant image feature information due to thepooling layers, capsule networks preserve more information by utilizing vector outputs that encode the probability and pose for an observation. By preserving pose information, capsule networks preserve the spatial relationships between features and are immune translations, rotations, and scaling transformations of image data. This approach was evaluated on the publicly available pneumoniaMNIST radiological dataset. The proposed method achieved a verification accuracy of 98.3%, which exceeds the performance of models such as ResNet-18, ResNet-50, autosklearn, AutoKeras, and Google AutoML.
Full Paper