A Comprehensive Framework for Lung Disease Classification using Luminosity-Guided Tubular Convolution

Abstract

This study presents an advanced methodology for multiple pulmonary disease detection including COVID-19 from CT images, combining a preprocessing normalization function for luminosity enhancement and a Convolutional Tubular Neighborhood (CNN) model. The proposed luminosity enhancement ensures consistent normalization, mitigating the impact of varied illumination conditions in CT scans and accentuating subtle features. The Convolutional Tubular Neighborhood CNN model, integrated into the feature extraction module, strategically focuses on tubular structures for precise information capture. Fine-tuned through 600 epochs with 256 neurons and a 5-Fold training scenario, our model achieved outstanding results, boasting an accuracy, specificity, sensitivity, and F1 Score of 0.99, 0.993, 0.99, and 0.99, respectively. The synergistic combination of luminosity enhancement and the specialized CNN architecture contributes to enhanced feature extraction, resulting in a robust and effective system for multiple pulmonary disease detection including COVID-19 detection. This methodology holds promise for diverse applications, particularly in healthcare, where accurate and efficient identification of COVID-19 from CT images is imperative.

DOI : https://doi.org/10.52783/pst.1740