A TRANSFER LEARNING APPROACH FOR PNEUMONIA DETECTION IN CT SCANS WITH SMOTE-BASED DATA BALANCING

Abstract

This research on using technology for critical health applications, specifically in detecting pneumonia through medical imaging, is truly impressive. This study focuses on the application of the deep learning Convolutional Neural Network (CNN) AlexNet model to classify lung Computed Tomography (CT) scan images and distinguish between normal and pneumonia-indicative conditions. This model is known for its simplicity and effectiveness in capturing intricate features from images. It has achieved remarkable performance on various benchmark datasets. In this research, comprehensive preprocessing techniques, including pixel rescaling and data augmentation, were implemented, along with addressing data imbalance using the Synthetic Minority Over-Sampling Technique (SMOTE). The findings are remarkable, with the developed AlexNet model achieving an impressive validation accuracy rate of 96.04% and 97.38% in classifying lung CT scans without and with SMOTE, respectively. The accuracy of this research is better compared to previous published research work. This highlights the potential of the AlexNet model as a reliable tool for pneumonia detection, leading to more efficient and accurate early diagnosis and timely treatment.