| MS Seminar

Name of the Speaker: Mr. B Jaya Chandra Raju (EE18S038)
Guide: Prof. Mohanasankar S
Online meeting link: https://meet.google.com/sjj-gcvi-vup
Date/Time: 10th January 2024 (Wednesday), 3:30 PM
Title: Missing Data Imputation of MRI Using Deep Learning

Abstract

MRI entails a great amount of cost, time, and effort for the generation of all the modalities that are recommended for efficient diagnosis and treatment planning. The advent of Machine Learning has led to the compelling need for data-driven solutions prompting large data initiatives that require the collation of MRI data from multiple scanning centers, often using different protocols. Retrospective harmonization minimizes these cross-site variabilities across the scans. Recent advancements in deep learning research show that GAN architectures have achieved substantial improvement in the aspects of style transfer and image synthesis. In this work, we formulate generating the missing MR modality from existing MR modalities as an imputation problem appended with the capability to transfer style for harmonization. The proposed model generated outputs that are closer to the target images and are on par with the state-of-the-art methods in terms of SSIM and PSNR. We show that harmonization is achieved by generating the output image incorporating the style code of the reference image. We further demonstrate that the proposed model exhibits a new degree of freedom, i.e., the ability to interpolate one single image both within and across modalities. Evaluation on the BraTS’15,18 multi-modal brain MRI datasets suggests that the method is promising and opens new avenues for further research.