.146 (MO.18) Automated segmentation analysis of nerve fascicular anatomy using high resolution ultrasound imaging for objective non- invasive assessment of nerve regeneration after VCA (Video of Presentation Available)
Thursday April 16, 2015 from 15:45 to 18:15
BRB Auditorium
Presenter

Vijay Gorantla MD, PhD, FRCS, United States

Associate Professor of Surgery | Administrative Medical Director

Reconstructive Transplantation | Department of Plastic Surgery

University of Pittsburgh Medical Center

Abstract

Automated segmentation analysis of nerve fascicular anatomy using high resolution ultrasound imaging for objective non- invasive assessment of nerve regeneration after VCA

Vijay Gorantla1, Vikas Shivaprabhu2, Howard Aizenstein3, John Galeotti2, George Stetten2.

1Plastic Surgery, University of Pittsburgh Medical Center, Pittsburgh, PA, United States; 2Bioengineering and Robotics , University of Pittsburgh and Carnegie Mellon University , Pittsburgh, PA, United States; 3Psychiatry, University of Pittsburgh Medical Center, Pittsburgh, PA, United States

Nerve Regeneration .

Background: Peripheral nerve (PN) regeneration after VCA is key to functional sensory motor outcomes. Currently, there is no non-invasive and economical imaging modality for sequential, reproducible monitoring of regeneration that correlates with validated measures and clinical functional outcomes after VCA.  As a key first step, we have used HRUS to successfully identify individual fascicles of normal nerves (eg. median n) and developed automated methods to reliably discriminate fascicles from other similar structures (such as tendons).

Methods: HRUS images of the normal median nerve were acquired in subjects (Visualsonics Vevo 2100 ®). Variance Descent Graphs (VDGs) were used to cluster pixels into homogeneous regions using a directed graph of edges between neighboring pixels. Each pixel forms a node in a graph, with a directed edge pointing to the neighboring pixel with lowest variance, assuming one exists with lower variance than itself.  Local minima in variance thus form the roots of disjoint trees.  The resulting trees each represent self-normalized relatively homogenous regions or patches that correspond to fascicles. These fascicular “patches” were then constructed, clustered and segmented using graph theory and 3 D reconstructed.

Results: Our algorithm detects expected shapes - circular regions - by making use of statistical data computed in robust neighborhoods of an image.  Our automated analysis segments circular structures by assignment of edge weights such that edges “within” a potential circular structure receive higher weights than “between” such structures. The analysis also ensures that meaningful results are obtained even in the presence of high noise.  Results presented for both 2D and 3D datasets confirm that the clustering algorithms employed are suitable to form fragments and perform the subsequent segmentation. 

Conclusion:. Our study confirms that HRUS can help identify fascicular anatomy of normal nerves.   Automated HRUS image segmentation can be helpful in differentiating “noise” (such as myelin debris, neuronal edema or axonal disruption) and provide contextual information of interest such as the growing axon cone of regenerating nerves in VCA. 

This study is funded by Department of Defense Grant - W81XWH-14-1-0370 . 3D Augmented High Resolution Ultrasound Imaging for Monitoring Nerve Regeneration and CR in VCA.


© 2017