Machine learning approaches for neurological disease prediction: A systematic review

In this article, we present a systematic and exhaustive review regarding the trends, datasets employed, as well as findings achieved in the last 11 years in neurological disorder prediction using machine learning models. In this work we present a comparison between the biomarkers used in ML field with the biomarkers that are obtained through other non-ml-based research fields. This will help in identifying the potential research gaps for ML domain. As the study of neurological disorders is a far-reaching task due to the wide variety of diseases, hence the scope of this study is restricted to the three most prevalent neurological diseases, that is, Alzheimer's, Parkinson's, and Autism Spectrum Disorder (ASD). From our analysis, it has been found that over time deep learning techniques especially Convolutional Neural Networks have proved to be beneficial for the disease prediction task. For this reason, Magnetic Resonance Imaging have been a popular modality across all three considered diseases. It is also notable that the employment of a transfer learning approach and maintenance of a global data centre helps in dealing with data scarcity problems for model training. The manuscript also discusses the potential challenges and future scope in this field. To the best of our knowledge, unlike other studies, this work attempts to put forth a conclusion of every article discussed highlighting the salient aspects of the major studies for a particular problem.