DEEP LEARNING-ENABLED LUNG DISEASE CLASSIFICATION USING A RESOURCE-EFFICIENT VGG ARCHITECTURE

Abstract

Deep learning has significantly advanced medical diagnostics by enabling accurate and automated detection of lung diseases. Its powerful feature extraction capabilities enhance early diagnosis, aiding in timely and effective treatment. This study focused on the identification of COVID-19, pneumonia, tuberculosis and normal cases through the utilization of an adopted lightweight model named VGG-12. This modified model based on convolutional neural networks provides a resource efficient and inexpensive solution for identification lung ailments at early stages. The model was trained on a large dataset of 14,000 images based on multi-class classification sourced from Kaggle. The adopted VGG-12 model obtained comparatively good results on the test dataset with 96.67% average accuracy with a reduced number of training parameters and training time. To improve the performance of the model and mitigate over fitting, advanced preprocessing and data augmentation techniques were used. The performance of the model was compared with multiple pre-trained networks like MobileNetV2, ResNet-50, and VGG-16. In comparison to these pre-trained models, the modified VGG-12 framework obtained comparatively good results in the classification of pneumonia, COVID-19, tuberculosis, and normal cases. This development will help medical professionals with a precise and quick diagnosis tool, contributing to timely and effective patient treatment. This work sets the stage for future advancements in automated medical diagnosis by demonstrating the potential of lightweight frameworks for efficient medical image processing, alongside the immediate benefits offered by the proposed model.