Multimodal Brain Image Registration for Neuro Mapping and Disorder Classification Using Multi-Pattern Kernel Learning

Brain disorders are accurately analyzed by integrating heterogeneous neuroimaging modalities in order to capture structural, functional, metabolic, and statistical features of the brain. The traditional unimodal techniques are often limited in their ability to fully characterize neurological complexity that leads to reduced diagnostic accuracy. To address this challenge, a novel Mapping and Classification via Multimodal Imaging-based Pattern learning Technique (MC-MIPT) is designed with multimodal neurological images. The proposed technique initially registers the multimodal data to spatially align input images. Subsequently, data pre-processing is also incorporated for artifact removal, noise reduction, and normalization to ensure clean and standardized inputs. The proposed MC-MIPT mainly designs a Multi-Pattern kernel learning Network that integrates patterned feature extraction, data fusion and a Deep learning (DL) network for neurological mapping and disorder classification. The Multi-pattern kernel learning machine is used to perform fine feature extraction for retrieving structural, functional, metabolic, and statistical features. The multimodal pattern fusion integrates the retrieved features for reducing dimensions and aligning cross-modalities. The complementary results are utilized by analyzing these processed fused features using a DL network for neurological mapping and disorder classification. In the experimental evaluation, the proposed MC-MIPT was validated with different metrics and achieved a superior classification accuracy of 95.7% with low processing time and very high scalability compared to state-of-the-art methods. Moreover, the proposed MC-MIPT improves the overall classification accuracy by 8.50%, 5.51%, 4.59%, 3.68% and 3.34% better than EIRAC, Hybrid CNN–GRU, GENet, GRP-DNet, and Ensemble Classifier.