3D Human Body Segmentation
Published in github.com, 2018
Cite with: Das, Neha. (2018). Development of a system that allows for the semantic segmentation of a 3D model of a human body into its constituent parts. http://neha191091.github.io/files/Semantic_Segmentation_IDP_Report.pdf
Abstract
Development of systems that are equipped to provide a view into a patient’s body form an essential part of the technological advancements needed to improve medical diagnosis and reduce invasive surgery. While volumetric imaging of a patient’s anatomy via Computed Tomography or Magnetic Resonance Imaging have achieved this goal to an extent, there is a growing need for combining such medical data via Augmented Reality Systems and Head Mounted Displays with the real-time view of the patient’s form. Applications of such an implementation include surgery planning and inter-operative guidance systems.
An essential component of an implementation of the aforementioned Medical Augmented Reality (MAR) System, would be a module that could segment patient’s form (obtained as a 3D Model) into its constituent body parts and identify the body part that corresponds to the volumetric medical data so that the two models may be registered.
The implementation of this segmentation module is the main goal of this InterDisciplinary Project. This is accomplished via first segmenting the depth maps obtained in the process of creating the patient’s 3D model using a Deep Convolutional Neural Network and then combining the labelled images via KinectFusion to obtain a segmented 3D Model.