Login | Register

Registration of Volumetric Ultrasound Images Using Siamese Neural Networks


Registration of Volumetric Ultrasound Images Using Siamese Neural Networks

Pirhadi, Amir (2021) Registration of Volumetric Ultrasound Images Using Siamese Neural Networks. Masters thesis, Concordia University.

[thumbnail of Pirhadi_MASc_S2022.pdf]
Text (application/pdf)
Pirhadi_MASc_S2022.pdf - Accepted Version
Available under License Spectrum Terms of Access.


In brain tumor resection, soft tissue deformation (i.e., brain shift) causes the pre-operative images to be invalid. Intra-operative ultrasound imaging is a non-invasive, portable, real-time, and cost-efficient alternative to track the surgery. A robust registration method is required to accurately align the post-resection and pre-resection ultrasound images for maximum tumor resection. However, registering the ultrasound images before and after resection is challenging for two main reasons. First, the tumor cavity after the surgery does not have correspondence in ultrasound images before the resection. Second, the quality of the image reduces during the surgery due to the blood clotting agents, air bubbles, and the saline water around the tumor.

This thesis proposes a robust non-rigid registration method based on a landmark tracking technique for brain shift correction in intra-operative ultrasound images. Some landmarks are selected in pre-resection ultrasound image manually that allows user-interaction. A Siamese neural network is adapted to track the annotated landmarks in the post-resection ultrasound image. The 2.5D approach enables 3D tracking and outlier detection. An optimal affine transformation is calculated using Iterative re-weighted least square (IRLS), which automatically suppresses the outliers.

The proposed method is tested on two publicly available datasets of REtroSpective Evaluation of Cerebral Tumors (RESECT) and Brain Images of Tumors for Evaluation (BITE). Mean target registration error (mTRE) is exploited for registration evaluation. In the BITE dataset, the method decreases the initial miss-alignment from 3.55±2.29 mm to 1.80±0.84 mm in pre/post-resection registration. In the RESECT dataset, mTRE is decreased from 3.55±1.76 mm and 3.49±1.56 mm to 1.26±0.57 mm and 1.12±0.46 mm in pre/post-resection registration and pre/during-resection registration, respectively. The fine-tuning effect is also assessed and is shown the generality of the method. The proposed method is compared to the state-of-the-art methods with statistical tests and showed average comparable or better results. The great accuracy, flexibility, and time-efficiency of the method make it an attractive option in real clinical applications that can increase the performance in neurosurgery.

Divisions:Concordia University > Gina Cody School of Engineering and Computer Science > Electrical and Computer Engineering
Item Type:Thesis (Masters)
Authors:Pirhadi, Amir
Institution:Concordia University
Degree Name:M.A. Sc.
Program:Electrical and Computer Engineering
Date:1 November 2021
Thesis Supervisor(s):Rivaz, Hassan and Ahmad, M. Omair
ID Code:989936
Deposited By: Amir Pirhadi
Deposited On:16 Jun 2022 15:01
Last Modified:16 Jun 2022 15:01
All items in Spectrum are protected by copyright, with all rights reserved. The use of items is governed by Spectrum's terms of access.

Repository Staff Only: item control page

Downloads per month over past year

Research related to the current document (at the CORE website)
- Research related to the current document (at the CORE website)
Back to top Back to top