Kyle Ewert
School of Applied Sciences, RMIT University, Melbourne, VIC, Australia
MSc Student

Kyle Ewert
MSc Student
School of Applied Sciences, RMIT University, Melbourne, VIC, Australia

Ryan L Smith
Radiation Oncology Medical Physicist
School of Applied Sciences, RMIT University, Melbourne, VIC, Australia; Alfred Health Radiation Oncology, Alfred Health, Melbourne, VIC, Australia

Max Hanlon
PhD Student
School of Applied Sciences, RMIT University, Melbourne, VIC, Australia

Rick Franich
Lecturer / Program Co-ordinator
School of Applied Sciences, RMIT University, Melbourne, VIC, Australia

Background

Due to the high doses involved in HDR Brachytherapy, incorrect delivery can result in imperfect treatment to the patient. Causes of incorrect treatment delivery can include catheter displacements due to tissue swelling. An approach that mitigates the risk of possible mis-deliveries is that of pre-treatment catheter localization using x-ray imaging and radio-opaque markers, to delineate the course of the catheters, allowing comparison with planned positions [1].

Purpose

Currently in the clinic, an A-P pre-treatment image is acquired using a flat panel detector (FPD) and the paths of catheter markers are manually marked for comparison with the TPS. This is a time-consuming process and potentially introduces inaccuracies. The aim of this investigation was to establish an automatic approach to locating the catheter markers and characterizing the displacement relative to the planned positions.

Methods

Pre-treatment images were acquired using overhead x-ray and FPD, of a solid water phantom containing fiducial markers and radio-opaque markers inserted into selected catheters. MATLAB code was used to perform contrast-limited-adaptive-histogram-equalization (CLAHE) to enhance feature-finding in a defined region of interest encompassing the implant. Initial `guess’ positions of the fiducial markers and small linear segments of markers were used to initiate the program, which then employs a region-growing process to accurately locate their positions and extent. The determined positions allow the program to automatically extrapolate to the next, unselected marker position. This iterative process determines the next marker locations using the prior marker positions and angles to predict the next position. When complete, the program returns the identified catheter channel number.

Results

Six sets of images were analysed: four phantom and two patient images. The process identified catheter paths as illustrated in figure 1. The extrapolation routine handled simple instances of catheter marker overlap and moderate curvature. A mean marker detection rate of 79% for the phantom images with no overlap, 72% with overlap and 29.5% for the patient images was achieved. Channel number identification was more difficult, with varying rates of success: 59% for the phantom images with no overlap, 30% with overlap, and is yet to be implemented for patient images.

Conclusions

This program demonstrates a method for automatically delineating radio-opaque catheter marker locations in the pre-treatment workspace, allowing for a reduced clinician time overhead. With the markers located, this potentially allows for automatic image registration and offset characterization, resulting in an easily obtained metric for determining the divergence of implant geometry from the plan.


← Back to all abstracts