Recent projects are being undertaken with Radiology, Cardiology, Neurology, Urology, Gynaecology, and more. We are keen to investigate projects in all clinical areas.
 

Real time Blood Flow Simulation

This projects aims to develop a set of tools for simulating human blood flow in real-time. The motivation is to provide framework that can be used in medical training and diagnostic simulators, where real-time response and accuracy are necessary requirements. The implementation is based on multiple smoothed particle hydrodynamics (SPH) models and leverages the power of modern GPUs using Nvidia's CUDA architecture. The flow model is independent of vascular anatomy topology and facilitates both synthetic and segmented models.
LeapMotion tool to measure joint flexibility 
Image of Leap Motion

Traditional methods for measuring metacarpophalangeal joint angles use the goniometer and can be time consuming.

The LeapMotion has the potential to reduce measurement time for thumb adduction, abduction and extension as well as to extract other data of interest.

Web technologies will be used to build rehabilitation software, designed to encourage hand and thumb exercises, as well as for assisting the practitioner monitor patient progress.
Prostate biopsy simulator
Image
Virtual training simulators for needle punctures reduce costs for theatre training and the risks to patients. Our development of a simulator for prostate biopsy addresses these issues. This will include producing a 2D graphical interface that closely matches the system used by the surgeon.

A Novint Falcon with an attached ‘Bard’ needle handle will be used to produce the haptic feedback during insertion of the biopsy needle.

A Phantom Omni will generate haptic feedback during the insertion of the ultrasonic transducer and will be used to control its orientation.
Interactive Segmentation of Multimodal Medical Data 
A novel interactive segmentation and editing tool will be developed, incorporating a simple user interface and a fast and reliable segmentation based on a 3D deformable model. Our interactive strategy allows the user to initialize the boundary by clicking a few control points on the desired object border and responds to the user in real time. The user can then edit the results by adding, removing, and moving the control points, where each interaction is followed by an automatic, real-time segmentation. During this process, the algorithm can dynamically learn the intension of the user based on the input priors (e.g., shape and appearance), and then predict the segmentation results for the user to further refine them. The accurate segmentation results can assist the surgeons in diagnosing diseases, including cardiovascular and brain problems.

VCath - a virtual learning iPad app to help train future neurosurgeons



A new mobile app, downloadable free of charge, will assist with the training of future neurosurgeons, and is just one of a stream of programmes being developed, adapting visual computing and three dimensional realities to provide cost-effective virtual learning for a range of medical procedures.

A basic skill that trainee neurosurgeons need to gain early in their training is an appreciation of the ventricular system in the brain, and how to cannulate it in an emergency. The flow of cerebrospinal fluid can be obstructed in the ventricles by many pathological processes leading to a dangerous condition known as hydrocephalus. The pressure within the ventricles can rise, leading to loss of consciousness.

The ventricular system can be cannulated in the operating theatre, fluid drained and the potentially lethal rise in pressure relieved. This procedure is a commonly performed in neurosurgical departments.

‘VCath’ enables trainees to improve their understanding and skill set needed for the procedure before having to perform in the operating theatre. Training using the cost effective simulator will help to provide neurosurgery trainees with the necessary skills for safe and effective completion of the procedure. It provides a convenient platform for repeat practice, anytime, anywhere, and has the potential to improve patient outcomes, safety and experience.

VCath is available to download, free of charge, directly from the Apple Store: Download VCath

Watch BBC Wales News Today news feature here.

Endoscope Simulator
 
The use of endoscope plays an important part in medical students’ training.  An endoscope can however remain only for a short time inside the patient’s orifice. Whilst employing the tool, it is therefore crucial for the trainee to learn how to operate it in a safe but fast manner.

The aim of this project is to develop an endoscope simulator that provides a safe and useful environment for medical students to acquire that skill. 

The endoscope simulator will be developed in C++ and use the open source 3D render engine Ogre. The endoscope will employ meshes that simulate environments of organs to explore through natural orifices including mouth and nostrils.


Intravascular Ultrasound Image Segmentation
Intra-vascular Ultrasound (IVUS) imaging is a catheter-based technology, which shows 2D cross-sectional images of the coronary artery. In this work, we present a shape prior based graph cut method for IVUS image segmentation, which does not require user initialisation. The shape prior is generalised from multiple training shapes, rather than using singular templates as priors. Weighted directed graph construction is used to impose geometrical and smooth constraints learned from priors. The proposed cost function is built upon combining selective feature extractors. A SVM classi er is used to determine an optimal combination of features in presence of calcification, brotic tissues, soft plaques, and metallic stent, each of which has its own characteristics in ultrasound images.


Left Ventricle Segmentation in SPECT
We propose a novel spatiotemporal constraint based on shape and appearance and combine it with a level set deformable model for Left Ventricle (LV) segmentation in 4D gated cardiac SPECT, particularly in the presence of perfusion defects. The model incorporates appearance and shape information into a soft-to-hard probabilistic constraint, and utilizes spatiotemporal regularization via a Maximum A Posteriori (MAP) framework. This constraint force allows more lexibility than the rigid forces of shape constraint-only schemes, as well as other state of the art joint shape and appearance constraints. The combined model can hypothesize defective LV borders based on prior knowledge.

Analysis of digital pathology cases with focus on muscle biopsy

Image

This project will seek to develop novel automatic analysis methods for use in digital pathology. 

Digital pathology is a growing field of research concerned with developing automatic methods to aid in the detection and diagnosis of pathological images.  Traditionally, a pathologist will use a microscope to visually assess a tissue biopsy for any pathological changes or problems.  However, with the advent of  high-quality slide scanners, and a national shortage of trained pathologists, there is a need to develop automated methods to assist pathologists in their daily work.

This project will initially look at developing novel methods for segmenting and analysing muscle biopsies with particular focus on developing robust and accurate segmentation algorithms.  

Breast Parenchymal Pattern Analysis In Digital Breast Tomosynthesis And Tabár Tissue


The aim of the project is to develop algorithms for mammographic segmentation based on Tabár tissue modelling.

With 2D mammographic projections (i.e. film and digital), one of the biggest challenges to screening radiologists is to interpret superimposed fibroglandular tissue (anatomical noise) in the images. It is because pathological structures can be obscured and remain undetected, in some cases this can mimic lesions leading to false positive results. This increases unnecessary recalls for additional screening using different modalities and/or biopsy if necessary. Recently advanced image acquisition using 3D digital breast tomosynthesis (DBT) shows improved tissue characterisation, tumour visualisation and strength in calcification localisation; leading to higher diagnostic performances and a lower recall rate for additional screening.

The use of 3D DBT has a significant implication of patient care; the clarity of the images provides advantages in mammogaphic interpretation while assessing the breast cancer risk. The proposed study is of interest in both scientific and clinical communities as the advanced technology results in images with fine tissue characterisation (opposite to superimposition as seen on mammography), which enables more accurate quantitatively measurement in parenchymal patterns using Tabár scheme. Therefore, to develop a technique that automatically segments a given mammographic image into regions, according texture and density variations can be found useful in quantification of change of relative proportion of different tissue, as means of aiding radiologists' estimation in mammographic risk assessment.The aim of the project is to develop algorithms for mammographic segmentation based on Tabár tissue modelling.

With 2D mammographic projections (i.e. film and digital), one of the biggest challenges to screening radiologists is to interpret superimposed fibroglandular tissue (anatomical noise) in the images. It is because pathological structures can be obscured and remain undetected, in some cases this can mimic lesions leading to false positive results. This increases unnecessary recalls for additional screening using different modalities and/or biopsy if necessary. Recently advanced image acquisition using 3D digital breast tomosynthesis (DBT) shows improved tissue characterisation, tumour visualisation and strength in calcification localisation; leading to higher diagnostic performances and a lower recall rate for additional screening.

The use of 3D DBT has a significant implication of patient care; the clarity of the images provides advantages in mammogaphic interpretation while assessing the breast cancer risk. The proposed study is of interest in both scientific and clinical communities as the advanced technology results in images with fine tissue characterisation (opposite to superimposition as seen on mammography), which enables more accurate quantitatively measurement in parenchymal patterns using Tabár scheme. Therefore, to develop a technique that automatically segments a given mammographic image into regions, according texture and density variations can be found useful in quantification of change of relative proportion of different tissue, as means of aiding radiologists' estimation in mammographic risk assessment.

Artefact reduction for pre-clinical CBCT imaging

Image of pre-clinical CBCT imaging

Metallic objects used in the monitoring of experimental subjects (such as ECG leads and thermometers) cause artefacts on conventional cone-beam CT reconstructions. The aim of this project is to remove / reduce these artefacts, rendering the reconstructions more useful for the purposes for which the data were acquired.

The artefact removal will be achieved by pre-processing the projection data prior to conventional FDK reconstruction (which may be carried out in commercial software). The pre-processing will use morphological and / or statistical analysis to determine the location and extent of the metallic objects in the projections, which will then be removed. This will be implemented in Python.

Application of projected spatial augmented reality for lumbar epidurals
This project aims to develop projected augmented reality technology to assist with lumbar epidurals by projecting the 2D X-Ray image of the spine onto the patients back. A prototype will display, on a screen, the lumbar CT scans overlaid onto the patient. Fiducial markers (e.g. marker pen cross-hairs) can be optically tracked to adjust the CT scans onto the image of the patient.

Projected spatial augmented reality (PAR) in surgery, where data is projected directly onto the patient skin, is exemplified by the work of Krempien et al and Sugimoto et al. This technology benefits the surgeon by removing the need for head mounted displays, or looking at augmented reality data on computer screens during surgery. It allows the surgeon to better focus on the patient and improve their awareness of the
underlying tissues.

The success rate for lumbar epidurals using palpation methods has been reported to be 30%.3 The low figure is attributed to; (1) inaccurate location of gaps between lumbar disks (2) needle trajectory are approximated and (3) difficulty in ‘feeling’ lumbar disks in patients with greater fat tissue.
Image
Simulation of cutting
Minimal invasive surgery is becoming increasingly popular, but the skills required by the surgeons are also becoming more demanding. Conventional approaches include using human anatomical models with the guidance of surgical consultants, or practising on real patients. The former is very costly and the latter places patients' lives at risk. To provide a more cost-effective and non-invasive solution, researchers at Cardiff's School of Computer Science are developing a image-guided system to train junior doctors and medical students in the basic surgical skills such as cutting. The system will provide users with a virtual environment that also includes real haptic feedback. The key differences between this work and previous work of artificially generating the virtual response is that the organ responses are supported by real mechanical models and the haptic forces are directly related to the real-time mechanical responses of human organs.
Image
 
Texture Based Analysis of Mammographic Images using Manifold Learning for Tissue Segmentation and Risk Assessment
Mammographic risk assessment is concerned with assigning a risk category to a given mammographic image based on the expectation of that patient developing breast cancer. The most commonly used risk assessment classification scheme is BI-RADS, which classifies mammograms into five risk assessment classes. Risk class one indicates a low risk of developing breast cancer, whereas risk class five indicates a high risk. The BI-RADS classification scheme is based on the density of the breast tissue, with a denser, more fatty, breast corresponding to a higher risk. As such, mammographic risk assessment models using the BI-RADS classification scheme will normally approach the problem by estimating and modelling the density of the breast to assign a risk assessment class to a given mammogram.