NeuroXNet: Creating a Novel Deep Learning Model Architecture that Diagnoses Neurological Disorders and Finds New Blood-Based Biomarkers with a miRNA Drug Discovery Pipeline Using Medical Imaging and Genomic Data
By Vaibhav Mishra
My research journey started from when I was a volunteer at a memory and rehabilitation center at my local city. There, I saw the drastic effects of neurodegenerative diseases on individuals making it incredibly hard to continue daily life. Motivated by my interest in neuroscience, I decided to start a research project in computational neuroscience … Neurological disorders continue to affect millions of people worldwide, with diseases leading to loss of cognitive function, a decline in memory, and even death. These diseases contribute to nearly a trillion dollars of healthcare spending and drastically change the lives of those affected. With the advent of new medical imaging and computational techniques, it has become possible to use large amounts of imaging data to build and train deep learning models that can diagnose many diseases with high accuracy rates using clinical tests and medical imaging tests like MRI. Some of the most common neurodegenerative disorders include Alzheimer’s disease, Parkinson’s disease, and Mild Cognitive Impairment … This study proposes a novel deep learning architecture, NeuroXNet, which performs multiclass diagnosis of AD, PD, MCI, glioma, meningioma, pituitary, and normal patients. NeuroXNet is the first model in published literature that diagnoses neurological diseases in seven classes using MRI images. This is also the first model in published literature which creates a novel architecture to classify neurodegenerative disorders instead of relying on previously built models like ResNet50 or VGG16. Furthermore, novel blood-based biomarkers and their corresponding miRNA regulatory pathways are identified with potential to aid in clinical drug discovery research through target identification, having the potential to drastically fasten the drug discovery process and reduce costs for in vitro experiments. In addition, NeuroXNet generates recommendations for treatment based on classification of disease from its convolutional neural network (CNN) model combined with the patient’s genomic data and clinical data …