Neuro-Geometric Graph Transformers with Differentiable Radiographic Geometry for Spinal X-Ray Image Analysis

Area of Science:

• Artificial Intelligence in Medical Imaging
• Neuro-geometric Deep Learning
• Explainable AI (XAI)

Background:

• Radiographic imaging interpretation is challenged by subtle visual cues, anatomical variations, and inter-observer variability.
• Conventional deep learning models, while predictive, often lack anatomical grounding and interpretability, hindering clinical trust.
• There is a need for AI systems that are both accurate and transparent in medical image analysis.

Purpose of the Study:

• To introduce SpineNeuroSym, a neuro-geometric imaging framework for explainable medical image analysis.
• To integrate geometry-aware learning and symbolic reasoning to improve trustworthiness and interpretability in spinal radiographs.
• To develop an AI system that addresses limitations of current deep learning approaches in medical imaging.

Main Methods:

• Developed SpineNeuroSym, a framework unifying weakly supervised keypoint/region discovery, a dual-stream graph-transformer, and a Differentiable Radiographic Geometry Module (dRGM).
• Incorporated a Neuro-Symbolic Constraint Layer (NSCL) for logical consistency and a Counterfactual Geometry Diffusion (CGD) module for validation and rare phenotype generation.
• Evaluated on 1613 spinal radiographs across six diagnostic categories: spondylolisthesis, infection, spondyloarthropathy, and normal cervical, thoracic, and lumbar spines.

Main Results:

• SpineNeuroSym achieved 89.4% classification accuracy, a macro-F1 score of 0.872, and an AUROC of 0.941 on the spinal radiograph dataset.
• The framework outperformed eight state-of-the-art imaging baselines in diagnostic performance.
• Demonstrated enhanced explainability, trustworthiness, and reproducibility through neuro-geometric modeling and symbolic constraints.

Conclusions:

• Integrating neuro-geometric modeling, symbolic constraints, and counterfactual validation advances explainable and trustworthy medical imaging AI.
• SpineNeuroSym establishes a pathway toward transparent and reproducible AI systems for medical image analysis.
• The framework shows significant potential for improving diagnostic accuracy and reliability in clinical practice.