Deep Learning Based Framework for DTI Parameters Estimation and Analysis for Sparse Diffusion MRI data

Abstract

Today, mental health issues are prevalent, which makes the diagnosis and prognosis of neurological newlinediseases crucial from a clinical perspective. Addressing mental health on a broad scale presents newlinechallenges in terms of both cost and time. Although psychiatrists typically address mental health newlinethrough therapy and counseling, the effectiveness of these approaches varies from person to person. newlineTherefore, noninvasive techniques such as diffusion tensor imaging (DTI) play a vital role in providing newlinequantitative measurements that assist in assessing mental health. Understanding the structure of white newlinematter is key to diagnosing and predicting mental health conditions accurately. Moreover, it is essential newlineto have quantitative measurements that are unbiased and can be deployed on a large scale. These DTI newlinequantitative parameters can be acquired in large scale in less amount of time using sparse diffusion newlineMRI. newlineSparse diffusion MRI, which can be acquired by small diffusion measurements, presents challenges due newlineto limited diffusion directions and inherent noise. Deep learning has emerged as a promising approach newlineto resolved these problem compare to traditional methods. Our thesis introduces a novel Deep Learning newlineBased Framework for DTI Parameter Estimation and Analysis tailored to sparse diffusion MRI data. newlineThis framework, incorporating Transformer Neural Network and Convolutional Neural Network (CNN), newlineaims to overcome the limitations of traditional DTI reconstruction methods. newlineWe conducted experiments on various datasets, including the Human Connectome Project (HCP) newlinewhich is high resolution, the MICCAI Quad22 Migraine dataset, the National Institute of Mental newlineHealth Data (NIFD), and the Alzheimer s Disease Neuroimaging Initiative (ADNI) which are mental newlinehealth diseases with lower resolution. Our findings show that our framework effectively improves DTI newlineparameter estimation and analysis for sparse diffusion MRI data. These results contribute to advancing newlineour understanding of brain connectivity and neurodegenerative diseases.