Skip to main content
Dr Emre Kopanoglu

Dr Emre Kopanoglu

Lecturer

School of Psychology

Email
kopanoglue@cardiff.ac.uk
Telephone
+44 (0)29 2251 0256
Campuses
CUBRIC 1.016, School of Psychology, Park Place, Cardiff, CF10 3AT
Users
Available for postgraduate supervision

Overview

News

We are recruiting! If you are interested in using Machine Learning and Magnetic Resonance Imaging, please consider applying to our studentship. For more information, please refer to the following link: https://www.findaphd.com/phds/project/epsrc-dtp-studentship-in-using-machine-learning-to-ensure-safety-of-patients-who-cannot-remain-still-during-magnetic-resonance-imaging/?p129512

Research summary

Magnetic Resonance Imaging is a powerful imaging modality with high soft-tissue contrast, inherent safety due to the lack of ionizing radiation, and diagnostically sufficient signal-to-noise ratio. My research aims to improve diagnostic image quality as well as patient comfort and safety in MRI, and involves signal/image processing, computer modelling, and novel imaging hardware.

With many MRI scans lasting several minutes, patient motion is a severe problem. If uncorrected in real-time, many motion patterns change the imaged volume, and therefore make imaging data inconsistent and necessitate re-scanning the patient. On the one hand, at lower field strengths, real-time (prospective) motion correction techniques can adapt the imaging volume in real-time. However, lower field strengths mean lower signal-to-noise ratio and contrast-to-noise ratio, i.e. lower image quality. On the other hand, ultra-high field (UHF, >3T) MRI offers many benefits in terms of image quality and contrast. Unfortunately, UHF MRI suffers from undesired contrast variations across the image. While such variations can be compensated for using tailored radiofrequency pulses and multi-channel transmit (parallel-transmit) systems, designing such pulses takes from upwards of several seconds to a few minutes with many algorithms. Therefore, real-time motion correction has not been possible yet with such pulses. My current research focuses on designing parallel-transmit pulses in real-time.

Biography

Education

  • 2012: PhD in Electrical and Electronics  Engineering. Bilkent University, Ankara, Turkey. Novel Techniques Regarding Specific Absorption Rate and Field of View  Reduction in Magnetic Resonance Imaging
  • 2006: BSc in Electrical and Electronics  Engineering. Bilkent University, Ankara, Turkey.

Employment

  • 2017 – present: Lecturer in Psychology. Cardiff University, Cardiff, UK.
  • 2015 – 2017: Senior Research Scientist. Aselsan Research Center, Ankara, Turkey.
  • 2012 – 2015: Post-Doctoral Associate. Radiology and Biomedical Imaging. Yale University, New Haven, CT, USA.
  • 2006 – 2012: Research and Teaching Assistant. Electrical and Electronics Engineering. Bilkent University, Ankara, Turkey.
  • 2006 – 2006: Undergraduate Teaching Assistant. Electrical and Electronics Engineering. Bilkent University, Ankara, Turkey.

Publications

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Current Research Interests 

My research focuses on Magnetic Resonance Imaging (MRI). More specifically, I am interested in safety and image quality in MRI.

Patient motion can cause patient heating to increase by more than 3-fold

An MRI scan causes patient heating. This heating is minimal and precautions are taken to ensure it stays below strict safety limits. For this purpose, computer simulations are performed using realistic body models (Figure 1). These computer simulations help us investigate the heating under realistic scan conditions. Then, we can adapt our imaging protocols to ensure safety.

Realistic body model image

Figure 1: Realistic body models are used in computer simulations to investigate patient safety. Here, the body model is shown together with a multi-channel transmit array. The latter is modelled as 8 rectangular loop coils.

One of the topics I investigate is the effect of patient motion on heating. MRI scans can often last up to an hour. Several patients, including children as well as adults, may have trouble staying still during such extended periods. Furthermore, patients with dementia, Parkinson’s, Tourette’s or Huntington’s may have tremors during MRI. When the patient’s position changes, the interaction of the scanner with the patient changes. As a consequence, the heating pattern may change (Figure 2).

Figure 2

Figure 2: A change in the patient’s position affects patient heating. Here, the hotspot moved from the anterior-left to the posterior-right part of the brain and it became hotter.

The Specific Absorption Rate (SAR) is related to patient heating, and is often used as a safety parameter in MRI. The variation of SAR across tissues is called local SAR. I design realistic radiofrequency pulses that would yield a high quality image. Then, I calculate the local SAR distribution using realistic body models. Finally, I investigate the change in local SAR in case of patient motion. Figure 3 depicts regions where local SAR exceeded the initially estimated peak due to patient motion; i.e., where heating would exceed the estimated maximum.

Figure 3

Figure 3: Patient motion cause the heating to more than double in a region with a volume of 19 cubic-centimetres.

Shorter scans can yield high quality images when images are processed together

In clinical settings, multiple imaging protocols are used to image a patient. These imaging protocols are adjusted such that each image set is under the influence of a different contrast mechanism (Figure 4). These images provide complementary information, and therefore, maximize diagnostic value.

Figure 4

Figure 4: Images acquired under the influence of different contrast mechanisms provide complementary diagnostic information.

To reduce scan time, MRI protocols can be accelerated by acquiring less data. If certain conditions are satisfied, the effect of this data reduction can be compensated for via image processing (Figure 5).

Figure 5

Figure 5: When the amount of data acquired is reduced by 87.5%, image quality is heavily affected (a). However, nonlinear reconstruction techniques can help us recover a high quality image (b).

When we are processing acquired data, we can process different contrasts together. This allows information sharing, and improves image quality (Figure 6). However, this joint processing may also cause detrimental effects. The most important such effect is the leakage of features that are unique to an image to the other images (leakage-of-features, Figure 6).

Figure 6

Figure 6: Processing images together (b) improves image quality compared to each image going through nonlinear reconstruction separately (a). However, this leads to leaking of features that are unique to one image to the other images (red arrows). Our proposed reconstruction method suppresses such leakage artefacts and yields artefact-free high-quality images (c). Please note that the image contrast was adjusted to maximize visibility of artefacts.

We proposed an image reconstruction algorithm that processes images both together and separately. Processing images together improves quality while processing images separately ensures that each image is faithful to its data. Therefore, the method yields high-quality images free of leakage-of-features (Figure 6). In-vivo images where the scan was accelerated by 87.5% show that high quality images can be acquired at 12.5% of the duration of a standard protocol (Figure 7).

Figure 7

Figure 7: Proton-density weighted, T1-weighted and T2-weighted images were processed together to reconstruct high quality images. All imaging protocols were 87.5% accelerated compared to their standard versions (acceleration factor R=8). The proposed method (SIMIT) showed the Lentiform Nucleus (pink arrows) and the frontal opercular cortex (yellow arrow) more clearly. SIMIT also depicted the gray-matter boundaries in the sulci more clearly in the T1-weighted images.

The proposed method (SIMIT) was evaluated by a neuroradiologist. The scores highlight the superior performance of SIMIT in terms of diagnostic value (Figure 8).

Figure 8

Figure 8: Neuroradiologist scores highlight the improved image reconstruction performance of SIMIT. The neuroradiologist was blinded to method names and images were presented in randomized order.

Research group

CUBRIC

Supervision

Current supervision

Luke Watkins

Research student

photograph of Alix Plumley

Alix Plumley

Research student