Multi-Class Classification of Pulmonary Diseases Using Computer Tomography Images

Abstract

The object of the study the architecture of a neural network designed for processing CT (computed tomography) scan images. The objective of this study is to examine the behavior of an existing neural network, originally constructed for binary classification, within the context of multi-class classification. To achieve this goal, two publicly available datasets were combined into a multi-class dataset, encompassing classes such as COVID-19, non-hospital pneumonia, and healthy lungs. The existing neural network architecture (ResNet50V2 with the utilization of the Feature Pyramid Network) was adapted for multiclass classification purposes. The resulting neural network was trained over 20 epochs, achieving an accuracy of 95.086% on a designated validation dataset. The experimental results demonstrate the potential of employing this and similar neural network architectures in the medical field, aiding healthcare professionals in their work. Ref. 7, pic. 3, tabl. 4

Keywords

• multi-class classification
• convolutional neural networks
• computer tomography image analysis