Advances in Multimodal Machine Learning for Brain Disorder Classification: A Comprehensive Review and Research Roadmap

Multimodal machine learning (MML) is a computational neurodiagnostic method that uses a combination of several data types: MRI, PET, EEG, clinical tests, and genomic data to provide a comprehensive picture of brain diseases. This review provides an overview of recent advancements in MML for important neurological disorders, such as Alzheimer's disease, Parkinson's disease, glioma, multiple sclerosis, schizophrenia, and epilepsy. Research studies consistently conclude that multimodal fusion models perform better than single-modality models, as they combine structural, functional, and molecular data, which enhances diagnosis and disease subtype classification. Nevertheless, there are still difficulties: there is no harmonization of data, a lack of generalization across locations, issues with interpretability, and inconsistent benchmarks. The new trends covered in the review include transformer-based fusion, self-supervised learning, and federated multimodal modeling, all of which aim to address the problems of data scarcity and privacy. Based on these findings, the proposed research will develop an interpretable, scalable MML architecture that combines imaging, clinical, and molecular data to allow for earlier diagnosis and more accurate patient stratification. The overall aim is to close the gap between algorithmic innovations and practical applications, increasing the validity and clinical adoption of multimodal neuroinformatics systems.