OrthoVector AI: A Physics-Guided Structure Tensor and Deep Learning Framework for Bone Fracture Screening on 2D X-Rays
Year: 2026
Type: preprint
Abstract: OrthoVector AI is an experimental biomedical research prototype designed to explore physics-guided feature extraction for fracture screening on standard 2D long-bone X-rays. Addressing diagnostic delays in resource-constrained primary care settings, the system combines classical biomechanical principles with deep learning. By applying local structure tensors grounded in Wolff’s Law, the engine analyzes directional trabecular alignment to highlight structural coherence loss and localized vector anomalies. These tensor metrics are fused with a fine-tuned ResNet-50 classifier outputs to generate a composite risk assessment. The framework includes explicit clinical limitation disclaimers, highlighting reduced reliability in complex joint spaces compared to long-bone shafts. This upload includes the technical manuscript outlining the mathematical foundation, dual-signal risk engine architecture, and empirical baseline validations.
Authors: Adhikari, Prince, Chaudhary, Ram Binay
Language: English
Topic: Medical Imaging and Analysis
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7212400008
Cites:
Cited by:
Related to: