Title: Q-Net: A Quantitative Susceptibility Mapping-based Deep Neural Network for Differential Diagnosis of Brain Iron Deposition in Hemochromatosis

Abstract: Brain iron deposition, in particular deep gray matter nuclei, increases with advancing age. Hereditary Hemochromatosis (HH) is the most common inherited disorder of systemic iron excess in Europeans and recent studies claimed high brain iron accumulation in patient with Hemochromatosis. In this study, we focus on Artificial Intelligence (AI)-based differential diagnosis of brain iron deposition in HH via Quantitative Susceptibility Mapping (QSM), which is an established Magnetic Resonance Imaging (MRI) technique to study the distribution of iron in the brain. Our main objective is investigating potentials of AI-driven frameworks to accurately and efficiently differentiate individuals with Hemochromatosis from those of the healthy control group. More specifically, we developed the Q-Net framework, which is a data-driven model that processes information on iron deposition in the brain obtained from multi-echo gradient echo imaging data and anatomical information on T1-Weighted images of the brain. We illustrate that the Q-Net framework can assist in differentiating between someone with HH and Healthy control (HC) of the same age, something that is not possible by just visualizing images. The study is performed based on a unique dataset that was collected from 52 subjects with HH and 47 HC. The Q-Net provides a differential diagnosis accuracy of 83.16% and 80.37% in the scan-level and image-level classification, respectively.