AI-driven deep CNN approach for multi-label pathology classification using chest X-Rays.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Artificial intelligence (AI) is crucial for chest disease detection and diagnosis through image analysis.
• Current AI methods struggle with identifying multiple chest diseases simultaneously due to data limitations and imbalance.

Purpose of the Study:

• To propose a synthetic data augmentation technique for improving the detection of 14 chest-related diseases.
• To evaluate the performance of deep Convolutional Neural Networks (CNNs) with augmented data for multi-class classification.

Main Methods:

• Employed three deep CNN architectures: DenseNet121, InceptionResNetV2, and ResNet152V2.
• Utilized synthetic data augmentation to address insufficient and imbalanced datasets.
• Trained and validated the models for classifying 14 distinct chest diseases.

Main Results:

• Achieved an average Receiver Operating Characteristic Area Under the Curve (ROC-AUC) score of 0.80.
• Demonstrated competitive performance compared to previous multi-class classification models for X-ray anomalies.
• The proposed model effectively classifies 14 chest-related diseases with enhanced accuracy.

Conclusions:

• Synthetic data augmentation significantly enhances AI performance in detecting multiple chest diseases.
• Deep CNNs, particularly DenseNet121, InceptionResNetV2, and ResNet152V2, show promise for accurate multi-class chest X-ray analysis.
• This research offers a valuable contribution to AI-driven healthcare for comprehensive chest disease diagnosis.