RVCNet: A hybrid deep neural network framework for the diagnosis of lung diseases

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Early diagnosis of lung diseases is crucial for reducing mortality.
• Computer-aided diagnostic systems (CADs) aid radiologists in accurate lung disease diagnosis.
• Classifying multiple lung diseases from radiographic images requires further research.

Purpose of the Study:

• To propose RVCNet, a hybrid deep neural network for multi-class lung disease prediction from X-ray images.
• To enhance feature extraction and classification using deep learning techniques.

Main Methods:

• Developed a hybrid framework (RVCNet) combining ResNet101V2, VGG19, and CNN models.
• Employed hyperparameter fine-tuning for feature extraction and added batch normalization, dropout, and dense layers for classification.
• Utilized a dataset comprising X-ray images of COVID-19, non-COVID lung infections, viral pneumonia, and normal cases (2262 training, 252 testing images).

Main Results:

• RVCNet achieved an overall classification accuracy of 91.27% with the Nadam optimizer.
• Performance metrics included AUC (92.31%), precision (90.48%), recall (98.30%), and F1-score (94.23%).
• RVCNet outperformed other standalone models like ResNet101V2 and VGG19 on this four-class dataset.

Conclusions:

• RVCNet demonstrates superior performance in classifying multiple lung diseases from X-ray images.
• The hybrid deep learning approach offers a promising tool for computer-aided lung disease diagnosis.
• GRAD-CAM visualization aids in interpreting RVCNet's image classification decisions.