Point cloud alignment in arm reconstruction for lymphedema detection by iterating pairwise ICP

Abstract

Arm lymphedema is a debilitating condition characterized by a large swelling of the patient arms. This thesis describes a system for measuring arm volume with the goal of providing a cost-effective, reliable, easy to use, comfortable, and informative alternative to the tedious water displacement method or the expensive perometer. The device consists of 8 inexpensive positionally fixed realsense sr300 depth cameras in a frame that can be placed on a flat surface, and used to scan a patient's arm in order to build a 3D mesh that can be used to extract not only the total volume of the arm but also the volume of specific segments. A special focus is given to the method of global registration of each camera's depth image, where camera positions are iteratively adjusted by applying ICP between one camera's point cloud and the point cloud generated by all-but-that-one camera. The system is then validated by taking a set of measurements of a human arm and comparing the values against those obtained from the water displacement method, in addition to also measuring objects of known volume and comparing those values with the ground truth. Despite a large non constant percentage error between expected and obtained values, early experiments show an 5_ correlation value of .99 between the expected and measured volumes. While additional work and experiments are needed to guarantee the reliability of this proposed method in order to be used in a clinical setting, these early findings show the potential of using such a system in small clinics where both space and money limits the accessibility of a large or expensive device. Keywords: lymphedema, global registration, realsense, point cloud alignment, volume measurement, 3D reconstruction