Loughborough University
Leicestershire, UK
LE11 3TU
+44 (0)1509 263171
Loughborough University

Loughborough University Institutional Repository

Please use this identifier to cite or link to this item: https://dspace.lboro.ac.uk/2134/5673

Title: Development of machine vision techniques for intraoperative registration and bleeding characterisation in robot-assisted neurosurgery
Authors: Gooroochurn, Mahendra
Keywords: Surgical robotics
Image processing and analysis
Pattern classification
Face processing
Colour segmentation
Image registration
Issue Date: 2009
Publisher: © Mahendra Gooroochurn
Abstract: The Mechatronics in Medicine Research Group at Loughborough University is developing an application of surgical robotics that will help the medical community resolve a pressing challenge in the management of head injuries. The medical community struggles to provide timely treatment in the aftermath of head injuries and due to its emergency nature, considerable effort and resources are spent in trying to deliver the treatment as soon as possible. Following an accident, the patient is taken to the closest hospital with Accident & Emergency (A&E) unit but not all such hospitals have a neurosurgical unit and therefore any neurosurgical intervention will not be possible without further transportation, which will result in detrimental effect in the health of the patient. In response to this shortcoming, the goal of the project is to design a robotic system, specifically a Mechatronic Intervention System for Emergency Neurosurgery (MISEN) that would have the necessary capabilities to aid any clinician to perform basic life saving emergency procedures at any A&E hospital without a neurosurgical unit. This thesis first presents the introductory research work carried out to analyse the medical aspects of the targeted neurosurgical procedures (Intracranial Pressure Monitoring, External Ventricular Drainage and Chronic Subdural Haematoma), from which a protocol for these three procedures and the engineering specifications of MISEN are set. The three neurosurgical procedures are broken down into basic tasks which are then mapped to engineering requirements, grouped as (1) Robotic Manipulator and Tools (2) Machine Vision Techniques (3) System Architecture and Control. The subsequent focus of the thesis is geared towards a preliminary investigation and design of two Machine Vision requirements. The first one allows the registration of a trajectory, defined in Computed Tomography (CT) scan frame of reference, to the robotic manipulator frame of reference. The second one is bleeding detection and characterisation which forms part of the robotic system supervisory protocol. The first Machine Vision technique relates to intraoperative registration, aimed at characterising the pose of the patient s head in the Operating Room using white light imaging. Intraoperative registration is a component of the whole registration framework; the other component is the characterisation of the preoperative space, which is performed in CT modality. Registration of the preoperative and intraoperative spaces allows the computation of an entry point on the patient s skull and a target point inside the cranium with respect to the robot coordinate frame of reference. Experimental tests and simulation studies are presented to validate the adequacy of using craniofacial landmarks (outer eye corners and ear tragi) as a registration basis to achieve a registration error within the required 5 mm accuracy. A point-based rigid body registration paradigm based on anatomical landmarks was used. Simulation studies, validated using experimental work on an artificial skull, show that an accuracy of 5 mm is achievable with the adopted technique. In view to furthering synergy between MISEN and the surgical team, automated extraction of the selected craniofacial landmarks was considered a desirable component of the system. The automated methods developed to extract these craniofacial landmarks were based on a neural network solution with Gabor features as inputs. The proposed feature detector and feature extraction technique were tested over a large number of images from publicly available face databases and found to yield good performance. The obtained high detection rate show that it is possible to automatically carry out the registration of the patient s head physical space to the preoperative CT scans. User validation will be introduced in the full implementation to achieve robustness and safe operation under all circumstances. The second Machine Vision technique developed is related to the detection of the onset of bleeding and give further information such as the location of the bleeding source(s) and the associated severity. These objectives were achieved by using image processing and analysis tools to extract features from flow propagation patterns. The results obtained provide a good foundation for performing further tests on components of the framework, such as blood segmentation on actual clinical materials (blood, tissue/skin/bone) and bleeding source(s) determination/tracking for patterns of blood flows obtained in practice.
Description: Closed access. A Doctoral Thesis. Submitted in partial fulfilment of the requirements for the award of Doctor of Philosophy of Loughborough University.
URI: https://dspace.lboro.ac.uk/2134/5673
Appears in Collections:Closed Access PhD Theses (Mechanical and Manufacturing Engineering)

Files associated with this item:

File SizeFormat
Mahendra Gooroochurn PhD Thesis.pdf14.41 MBAdobe PDFView/Open

 

SFX Query

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.