Neuro-Bridge-X: A Neuro-Symbolic Vision Transformer with Meta-XAI for Interpretable Leukemia Diagnosis from...

Area of Science:

• Artificial Intelligence in Medicine
• Computational Pathology
• Machine Learning for Hematology

Background:

• Acute Lymphoblastic Leukemia (ALL) diagnosis is challenging due to subtle symptoms and invasive traditional methods.
• Bone marrow biopsies and flow cytometry are costly, time-consuming, and invasive procedures.
• There is a need for non-invasive, automated, and explainable diagnostic tools for ALL.

Purpose of the Study:

• To develop and evaluate Neuro-Bridge-X, a novel neuro-symbolic hybrid AI model for automated ALL diagnosis.
• To utilize peripheral blood smear (PBS) images for accurate and explainable ALL detection.
• To enhance diagnostic precision and clinical workflow efficiency in oncology.

Main Methods:

• Neuro-Bridge-X integrates deep morphological feature extraction and vision transformer-based contextual encoding.
• Fuzzy logic-inspired reasoning and adaptive explainability are key components of the model.
• The model was trained and validated on two large datasets (ALL Image and C-NMC) using advanced data augmentation and cross-validation techniques.
• Evaluated optimizers include Nadam, SGD, and Fractional RAdam.

Main Results:

• SGD optimizer achieved near-perfect accuracy (1.0000 on ALL, 0.9715 on C-NMC) with robust generalization.
• Fractional RAdam also demonstrated high performance (0.9975 on ALL, 0.9656 on C-NMC).
• Nadam optimizer showed inconsistent convergence and lower accuracy on the C-NMC dataset.
• A Meta-XAI controller provided dynamic selection of explanation strategies (Grad-CAM, SHAP, Integrated Gradients, LIME) for interpretability.

Conclusions:

• Neuro-Bridge-X, particularly with SGD and Fractional RAdam optimizers, accurately identifies ALL by focusing on critical morphological features.
• The model provides a scalable and interpretable solution for ALL diagnosis, surpassing limitations of conventional methods.
• This AI-driven approach has the potential to significantly improve diagnostic accuracy and streamline clinical workflows in hematological oncology.