Abstract
INTRODUCTION: Single-photon emission computed tomography/computed tomography (SPECT/CT) imaging plays a critical role in sports injury diagnosis by offering both anatomical and functional insights. However, traditional SPECT/CT techniques often suffer from poor image quality, low spatial resolution, and limited capacity for integrating multiple data sources, which can hinder accurate diagnosis and intervention. METHODS: To address these limitations, this study proposes a novel multimodal learning framework that enhances SPECT/CT imaging through biomechanical data integration and deep learning. Our method introduces a hybrid model combining convolutional neural networks for spatial feature extraction and transformer-based temporal attention for sequential pattern recognition. This study further incorporates a biomechanics-aware injury detection module (BID-Net), which leverages kinematic signals, motion data, and physiological context to refine lesion detection accuracy. RESULTS: Experimental results on a curated sports injury dataset demonstrate that our framework significantly improves image clarity, diagnostic precision, and interpretability over traditional approaches. DISCUSSION: The integration of biomechanical constraints and adaptive attention mechanisms not only enhances SPECT/CT imaging quality but also bridges the gap between AI-driven analytics and clinical practice in sports medicine. Our study presents a promising direction for intelligent, real-time diagnostic tools capable of supporting injury prevention, early detection, and rehabilitation planning in athletic care.