DPU - Symposium 2022

Medica l Image Ana lysis & Ar t i f icia l Intel l igence Symposium 2022 Nr. 18: S. M. Ragib Shahriar Islam: Learning Cytoarchitectonic Structure From 3D Polarized Light Imaging Curriculum Vitae: Mr. S. M. Ragib Shahriar Islam is a Ph.D. researcher, working at ACMIT Gmbh. He is currently working on the development of optimization techniques for the reconstruction of CBCT images. He has completed his Master’s study in an Erasmus Mundus Joint Master’s Degree program entitled “Medical Imaging and Applications (MAIA)”, in a collaboration with the University of Burgundy (France) for semester-1, University of Cassino and Southern Lazio (Italy) for semester-2, University of Girona (Catalonia, Spain) for semester-3, and Forschungszentrum Jülich (Germany) for master’s thesis. He completed his undergraduate studies in Electrical and Electronic Engineering. His research interests include Biomedical Engineering, Computer Vision, Medical Image Acquisition and Analysis Techniques, Machine and Deep Learning, Generative Adversarial Networks (GANs), and Robotics. Abstract: Learning Cytoarchitectonic Structure From 3D Polarized Light Imaging 3D Polarized Light Imaging (3D-PLI) is one of the most successful procedures for revealing the nerve fibers’ organization inside post-mortem brain microscopy samples. Using the birefringence property presented in the myelin sheaths, the polarization microscope can measure the nerve fiber orientation at the sub-millimeter level by passing a polarized light beam through the brain sections without requiring any staining. Therefore, ideally suited for multi-modal analysis by combination, e.g., with staining for cell bodies after the 3D-PLI measurement. Nevertheless, the acquisition process of such multi-modal data is challenging and time-consuming, enabling access to only a limited number of samples. This study aimed to find the cytoarchitectonic features inherent in the 3D PLI data, circumventing the costly data acquisition, using scalable unsupervised deep learning methods. The parameter maps of 3D PLI images were used to predict the cytoarchitectonic image following a progressive process from UNET to region mutual information-based conditional GAN.