CROSS-MODEL SPATIAL-SEMANTIC FUSION: INTEGRATING MRI AND HISTOPATHOLOGY FOR AUTOMATED BONE-SARCOMA SEVERITY ASSESSMENT

Abstract

The accurate diagnosis and severity staging of bone-sarcoma inherently depend on synthesizing macroscopic radiological imaging (MRI) with microscopic histopathological analysis. However, contemporary deep learning frameworks process these modalities in strict isolation, creating a diagnostic bottleneck that limits clinical utility. In this paper, we propose the Cross-Modal Spatial-Semantic Diffusion Network (CM-SSD), a novel multi-modal fusion architecture. CM-SSD synchronously processes macroscopic tumor dissemination from MRI alongside microscopic malignancy patterns from Hematoxylin and Eosin (H&E) slides. By leveraging dual spatial-semantic encoders and a cross-modal attention mechanism, the framework adaptively aligns complementary features across both modalities. This integrated approach not only enhances automated tumor localization but also generates a quantitative, percentage-based severity score. By bridging the gap between radiological and pathological computer-aided diagnostics, CM-SSD provides a unified, highly robust framework for real-time clinical decision support.