An international conference connecting people
in CAD research, education and business
Bookmark and Share
Copyright (C) CAD Solutions, LLC. All rights reserved.
Proceedings of CAD'17, 2017, 478-481
Towards Integration of Algebraic Topological Methods in Deep Learning from Medical Images

Francesco Furiani, Enrico Marino, Federico Spini, Alberto Paoluzzi, Roma Tre University

Abstract. We discuss here the introduction of basic algebraic topological methods, specifically linear spaces of chains and cochains, and linear operators of boundary/coboundary and their combinations, in 3D medical image segmentation, using state-of-the-art software and hardware. In this abstract we describe how we started using a deep learning approach for performing 3D medical image segmentation, by using unsupervised system identification of analytical models and algebraic topology methods with. In our research we use "chains" of image cells as representations in neural network (NN), and employ sparse matrices of boundary and coboundary operators to discriminate between statistical neighborhoods of various portions of body organs. The research is within the framework of MedTrain3dModSim, an EU project in the Erasmus+ program, where an international group of researchers from Europe and Korea started preparing CAD tools for training of students of medical schools within an international curriculum, using medical 3D modelling for education and treatment. This endeavor will make use of 3D printed models of body organs and virtual and augmented reality tools, in the perspective of going towards environments of modeling and simulation for computer-aided surgery. This project strongly relies on the IEEE project for standardization of model extraction from 3D medical images.

Keywords. Deep Learning, Medical Imaging, Segmentation, Model Extraction, LAR

DOI: 10.14733/cadconfP.2017.478-481