Convolutional Neural Networks architectures for lung classification

O. E Makinde; D.A. Omideyi; M. O Oyediran

Convolutional Neural Networks architectures for lung classification

Abstract: The increasing prevalence of lung-related diseases such as COVID-19, pneumonia, and tuberculosis has necessitated the development of intelligent diagnostic systems for early detection and classification using Chest X-ray images. This study focused on the development of a deep learning-based lung disease classification framework using Convolutional Neural Networks (CNN) integrated with the Mayfly Optimization Algorithm (MA). This study objectives are to: (i) acquire and pre-process Chest X-ray image datasets for lung disease detection, (ii) perform image segmentation and feature extraction using CNN architecture, (iii) optimize CNN hyper-parameters using the Mayfly Algorithm, and (iv) evaluate the performance of the developed CNN-MA model in comparison with the conventional CNN classifier. The study adopted deep learning techniques for lung disease classification using Chest X-ray images. A total of 2,820 Chest X-ray images comprising 1,414 COVID-19 positive and 1,406 non-COVID images were obtained from publicly available medical image repositories. The dataset was partitioned into 60% training and 40% testing sets using random sampling cross-validation. Data preprocessing techniques such as image resizing, histogram equalization, augmentation, segmentation, and feature extraction were performed. The CNN architecture was integrated with the Mayfly Algorithm for hyper-parameter optimization and feature selection. Performance evaluation was conducted using False Positive Rate (FPR), Sensitivity, Specificity, Accuracy, and Recognition Time. This findings revealed that: (i) the conventional CNN classifier achieved an accuracy of 90.16%, sensitivity of 89.96%, specificity of 90.36%, and FPR of 9.64%; (ii) the developed CNN-MA classifier achieved an improved accuracy of 96.19%, sensitivity of 95.95%, specificity of 96.43%, and FPR of 3.57%; (iii) the CNN-MA classifier recorded a faster recognition time of 89.54 seconds compared to 101.00 seconds for the conventional CNN; and (iv) the optimal CNN-MA configuration consisted of 3 convolutional layers, 128 filters, filter size of 6×6, and batch size of 155. This study concluded that the integration of the Mayfly Optimization Algorithm with CNN significantly improved lung disease classification performance by enhancing feature optimization, reducing false positive rates, and improving classification accuracy. This study recommended the adoption of intelligent MA-CNN frameworks for automated lung disease diagnosis in healthcare systems. However, this study was limited to publicly available Chest X-ray datasets and binary classification of lung diseases.

KJSET Volume. 5, Issue 1 (2026)

Contributor(s)

Keywords

Download Full-text (PDF)

Convolutional Neural Networks architectures for lung classification