Abstract:
Accurate and efficient classification of lung diseases from chest X-ray (CXR) images is
vital for timely diagnosis and treatment, particularly in healthcare environments with
limited resources. This study introduces LightXrayNet, a lightweight convolutional
neural network designed for multi-class classification of CXR images into nine
categories: Normal, Pneumonia, Higher Density, Lower Density, Obstructive Pulmonary
Diseases, Degenerative Infectious Diseases, Encapsulated Lesions, Mediastinal
Changes, and Chest Changes. The dataset was sourced from the publicly available X-ray
Lung Diseases Images (9 Classes) repository on Kaggle and subjected to a
comprehensive preprocessing pipeline, including adaptive CLAHE-based contrast
enhancement, resizing, normalization, light augmentation (horizontal flip, ±5° rotation),
and data splitting. LightXrayNet’s performance was benchmarked against three
pretrained CNNs—DenseNet201, ResNet50V2, and InceptionV3—using metrics such
as accuracy, precision, recall, F1-score, confusion matrices, and training efficiency.
Experimental results show that LightXrayNet achieved a test accuracy of 99.22% and
near-perfect values across all classes, while requiring substantially less training time
compared to deeper pretrained architectures. These findings demonstrate the potential of
LightXrayNet as a practical and deployable solution for automated lung disease
detection, with strong applicability in resource-constrained healthcare settings.
