Computer-aided diagnostic for classifying chest X-ray images using deep ensemble learning.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Computer-Aided Diagnosis

Background:

• Radiologists face heavy workloads, necessitating efficient diagnostic tools.
• Computer-Aided Diagnosis (CAD) systems leverage AI for faster, more accurate diagnoses.
• Deep learning offers significant potential for developing advanced CAD systems.

Purpose of the Study:

• To propose a novel Convolutional Neural Network (CNN) ensemble architecture.
• To classify chest X-ray (CRX) images into viral Pneumonia, Tuberculosis, COVID-19, and Healthy categories.
• To enhance early-stage diagnosis and treatment accuracy.

Main Methods:

• Utilized transfer learning and data augmentation techniques.
• Designed and evaluated two CNN ensemble models: Stacking and Voting.
• Integrated six diverse base CNN models (VGG16, VGG19, InceptionV3, ResNet101V2, DenseNet121, CheXnet).

Main Results:

• Achieved 99% accuracy with the Stacking Ensemble model.
• Attained 98% accuracy with the Voting Ensemble model.
• Demonstrated superior results and generalization compared to previous works.

Conclusions:

• The developed CNN ensemble system is ready for application in CAD systems for automated diagnosis.
• The approach provides a valuable second opinion for medical professionals.
• Initiated explainable deep learning to offer further insights for CRX evaluation.