Multi-scale, multi-parametric data acquisition and analysis for healthcare.
Our research can be described under the following headings:
Magnetic resonance imaging and spectroscopy (MRI/S)
Magnetic resonance spectroscopy and imaging holds the promise to identify specific metabolites using techniques from quantum control which are commonly used in Nuclear Magnetic Resonance (to identify chemical composition, molecule structures, etc.)
However, practical considerations of imaging and spectroscopy in patients (complex biological environments at room temperature compared to the highly controlled environments of Nuclear Magnetic Resonance) cause many uncertainties which makes it hard to obtain reliable data or any usable data at all.
Recent approaches from robust quantum control deep learning and advances in sensing technology help to improve these techniques to obtain more reliable information about specific metabolites. This in turn can help to identify biomarkers for early diagnosis of diseases such as cancer and dementia. It also builds the basis to obtain quantified data to build functional models of biochemical processes.
We are developing control techniques to compute custom magnetic resonance imaging and spectroscopy pulse sequences, based on user-defined targets, such as quantifying specific metabolites. Quantification of the data acquired with such techniques is done using traditional spectral analysis techniques as well as deep learning.
Medical image segmentation
We work on medical image segmentation using deep learning. We work on techniques for brain tumour and stroke lesion segmentation and develop techniques to detect early signs of Alzheimer's disease based on MRI, CT and PET images. We further devise methods for computer-aided diagnosis of early-stage prostate cancer and lung cancer diagnosis and treatment planning based on CT.
Human task performance
We aim to develop computational models that can automatically and reliably predict the task performance of the radiologist in the interpretation (such as lesion detection) of medical images. These models will be used either to support the human to augment diagnostic efficiency, or to train the human towards improved diagnostic accuracy.
Cwrdd â'r tîm
- Yang, J. et al., 2020. Self-paced balance learning for clinical skin disease recognition. IEEE Transactions on Neural Networks and Learning Systems 31 (8), pp.2832-2846. (10.1109/TNNLS.2019.2917524)
- Lévêque, L. , Young, P. and Liu, H. 2020. Studying the gaze patterns of expert radiologists in screening mammography: a case study with Breast Test Wales. Presented at: 28th European Signal Processing Conference (EUSIPCO 2020) Amsterdam, Netherlands 18-22 Janurary 2021.
- Jenkins, C. et al. 2019. Quantification of edited magnetic resonance spectroscopy: a comparative phantom based study of analysis methods. Presented at: ISMRM 27th Annual Meeting & Exhibition Montréal, QC, Canada 11-16 May 2019. , pp.-.
- Yu, Y. et al., 2018. Perceptual quality and visual experience analysis for polygon mesh on different display devices. IEEE Access 6 , pp.42941-42949. (10.1109/ACCESS.2018.2859254)
- Leveque, L. et al. 2018. State of the art: Eye-tracking studies in medical imaging. IEEE Access 6 , pp.37023-37034. (10.1109/ACCESS.2018.2851451)
- Al Ali, A. et al., 2015. The influence of snoring, mouth breathing and apnoea on facial morphology in late childhood: a three-dimensional study. BMJ Open 5 (9) e009027. (10.1136/bmjopen-2015-009027)
- Kajić, V. et al., 2010. Robust segmentation of intraretinal layers in the normal human fovea using a novel statistical model based on texture and shape analysis. Optics Express 18 (14), pp.14730-14744. (10.1364/OE.18.014730)
- Legg, P. A. 2010. Multimodal retinal imaging: Improving accuracy and efficiency of image registration using Mutual Information. PhD Thesis , Cardiff University.