Deep Learning (DL)-based spinal cord injury (SCI) prediction is an unswerving model in medical imaging. However, the prediction process is challenging during the segmentation and classification process. Conventionally, radiologists examine the spinal cord images to indicate the abnormalities in the spinal cord manually. The manual high dimensional spinal cord feature space interpretation makes it complex to identify the severity level. However, DL approaches help in faster and more accurate predictions. The model intends to predict the abnormal and normal spinal cord images automatically. Under certain conditions, the weight of the training images is based on similarity distribution which is likely to show better performance. It does not deal with unrepresentative data. Thus, the kernel-based learning process intends to reduce the difference between the testing and training data and explore the kernel learning value for image weighting. Here, a novel kernel-based weighting method reduces the maximal mean discrepancy (MMD) among the testing and training data, facilitating kernel and weighted image optimization. Experimental results demonstrate that the anticipated kernel-based image weighting model has a higher computational disorder prediction ability than other approaches. The numerical outcomes prove that the model has superior performance to the prevailing approaches regarding prediction accuracy, recall and precision.