Abstract
Abstract Cervical spondylosis (CS) is a progressive degenerative spinal disease in which its late or inaccurate diagnosis can result in permanent neurological brain injury and a high clinical demand of automated and stage-sensitive diagnostic devices. Nevertheless, existing AI vectors of spine diagnosis have severe shortcomings, such as biases with the static data, loss of high-frequency pathological signalization, absence of multi-planar feature combinations, and poor connections of imaging results with functional clinical signs. In order to fill these gaps, this paper will present Inception ResNet Layers, Dual Image Multi-layer Mapping, and an Auto-Decision-Making algorithm as a way of early cervical spondylosis diagnosis, classification, and severity. The model combines spinal color images (SCI), black and white spinal images (SBW), and patient clinical symptom information to facilitate multimodal and hierarchical clinical inference. The approaches include exponential decomposition-based feature extraction, dual-stream convolutional learning, hierarchical dual-label encoding, and multi-task learning to classify and regress the severity simultaneously. Stage-based adaptive optimization on Adam and on momentum scheduling boosts the convergence and generalization. Experimental findings indicate high-quality early detection, high true positive and negative classification rates, high diagnostic confidence, and better consistency in predicting severity than the use of single-modality and traditional deep learning systems does. The suggested model has reached up to 96.8% accuracy of dual-image fusion with a considerable decrease in false negatives, which shows great possibilities for future clinical screening and AI-based decision support systems.