YOLOv8 framework for COVID-19 and pneumonia detection using synthetic image augmentation.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Deep Learning for Diagnostics

Background:

• Accurate detection of COVID-19 and pneumonia via medical imaging is crucial for patient management.
• Dataset imbalance and lack of interpretability hinder AI adoption in diagnostics.
• Explainable AI (XAI) techniques are needed to build trust in AI-driven medical diagnoses.

Purpose of the Study:

• To develop a robust framework integrating synthetic image augmentation and deep learning (DL) for improved COVID-19 and pneumonia detection.
• To address dataset imbalance and enhance diagnostic accuracy using advanced AI models.
• To increase trust in AI diagnoses through Explainable AI (XAI) techniques.

Main Methods:

• Benchmarking state-of-the-art DL models (InceptionV3, DenseNet, ResNet).
• Generating synthetic images using Feature Interpolation and PCA for data augmentation.
• Training YOLOv8 and InceptionV3 models on augmented data, utilizing Grad-CAM for explainability and LLMs for analysis.

Main Results:

• YOLOv8 achieved 97% accuracy, precision, recall, and F1-score, surpassing benchmark models.
• Synthetic data generation effectively mitigated class imbalance and improved recall for minority classes.
• XAI visualizations (Grad-CAM) confirmed model focus on clinically relevant areas, validating diagnostic decisions.

Conclusions:

• The integrated framework enhances COVID-19 and pneumonia detection, fostering trust via synthetic data, DL, and XAI.
• High accuracy of YOLOv8 combined with interpretable visualizations promotes clinical AI adoption.
• Future work includes developing a human-in-the-loop workflow and integrating transformer models for enhanced interpretability.