Student Profiles

2019 Cohort

Thomas Allcock
Research Topic: Explainable AI in Digital Pathology: The Role of Visualising Deep Neural Networks in Supporting the Augmented Pathologist
Supervisors: Dr Rebecca Brannan, Dr Andy Hanby, Professor Andy Bulpitt
About Tom: Tom completed undergraduate and masters degrees in Physics at Nottingham.
Project Description: Breast cancer is the most common cancer in the UK, affecting 54,500 women in 2016. The diagnosis of breast cancer is performed by pathologists who are under pressure with an increase in the volume and complexity of the work, and a lack of qualified doctors available to do it. Digital pathology offers a solution, by allowing pathologists to use a computer to view pathology slides and the possibility of using artificial intelligence to improve the diagnostic accuracy and speed of breast cancer diagnosis. In order for humans to trust such methods, we need to address their explainability. This project aims to create diagnostic models that are able to summarize the reasons for behaviour, gain the trust of users, and produce insights about the causes of their decisions when diagnosing breast cancer.
Andrew Broad
Research Topic: Can attention-like Mechanisms help in a Semantic understanding of Imaging Data?
Supervisors: Dr Darren Treanor, Dr Marc de Kamps
About Andrew: Andrew graduated in Electronic Engineering at Leeds and retrained as a software engineer. Most recently he worked at LTHT, in the team developing the award-winning patient record system, PPM+.
Project Description: If a patient is suspected of having cancer, a tissue sample may be taken for further investigation. This biopsy is processed and scanned to yield a high-resolution digital image which is examined by a histopathologist. My PhD project will explore whether Artificial Intelligence can support the pathologist in reaching a diagnosis. The problem is that existing AI algorithms would struggle to process such large images quickly and accurately. This project will examine attention in artificial neural networks, to see whether emulating visual attention processes in the human eye and brain can identify areas of the biopsy image that are most likely to contain cancer cells. These regions can then be examined in greater detail, to deliver a more accurate diagnosis more rapidly.
Lucy Godson
Research Topic: Predicting Cancer Outcomes through integration and analysis of molecular, cellular and clinical data
Supervisors: Dr Ali Gooya, Professor Graham Cook, Dr Navid Alemi, Dr Jeremie Nsengimana (University of Newcastle)
About Lucy: Lucy achieved a 1st class honours degree in Biochemistry with industrial experience at the University of Manchester.
Project Description: Malignant melanoma has a poor prognosis for patients and presents as a highly heterogeneous disease. Therefore, it is hoped that through using different deep learning techniques more can be understood about the disease progression and how it affects different individuals. The project aims to combine medical imaging, genetic, clinical and lifestyle data to gain better biological insight of this complex disease and be able to stratify patients into treatment groups.
Rachael Harkness
Research Topic: Deep learning based Lung Nodule Detection and Cancer risk stratification through the effective Integration of Imaging and Electronic Medical Records (EMRs)
Supervisors: Professor Geoff Hall, Professor Alex Frangi
About Rachael: Rachael graduated with a BSc Hons in Biomedical Science and an MSc in Bioinformatics from Newcastle University.
Project description: Lung cancer is the leading cause of cancer related deaths in the UK with very low five- and ten-year survival rates. This is attributed to the fact that the cancer is typically diagnosed at an advanced stage as early detection is particularly challenging. Consequently, there is a clear need for automated and robust systems that facilitate its early detection, diagnosis and treatment. The goal of this project is to develop a system to improve patient management in the context of both lung cancer screening and incidental nodule discovery. The project objectives are defined to this end: · Develop a fully supervised lung nodule detection framework for automatic assessment of low dose chest CTs. · Develop a cancer risk stratification system based on the official clinical guidelines. · Investigate the use of NLP tools for extracting useful information from electronic medical records (EMRs), for subsequent integration with the image analytics. · Develop a weakly supervised nodule detection algorithm for the effective integration of NLP-derived data from radiology reports with imaging data (low-dose chest CTs).
Sara Jones
Research Topic: Using Behavioural AI for the Early Detection of Brain Cancer
Supervisors: Dr Ryan Matthews, Dr He Wang
About Sara Sara graduated with a BSc in Computing from the University of Worcester in 2018 before pursuing an MSc (Eng.) in Mechatronics and Robotics at the University of Leeds.
Project Description This PhD project aims to explore the feasibility of a custom behavioural AI tool as an early detector of a brain tumour. We propose that since brain cancer is a solid, structural lesion affecting a person’s most functionally eloquent organ, its presence will cause differences in language processing and cognition, and that this can be detected using AI techniques that are adequately trained to detect changes in personality, cognition, language processing (including voice and facial expressions) and prosodic and linguistic features of the patient’s speech. Using natural language processing techniques and Scaled Insights core neural net technology, survey answers, audio samples and transcripts will be analysed to generate markers pertaining to sample syntax, grammar, cohesion, discourse and content described.
Jason Keighley
Research Topic: Can Deep Learning work on Reduced Data Sets? A Case Study on histopathological Data.
Supervisors: Dr Darren Treanor, Alexander Wright, Dr Marc de Kamps
About Jason Jason completed a BSc in Computer Forensics and Security at Leeds Beckett University and achieved a distinction in an Advanced Computer Science masters at the University of Leeds.
Project description Digital pathology slides have very high resolutions, increasing neural network layer sizes and training times, with patch learning reducing image context. My research aims to focus on digital pathology slide dimensional reduction using Variational Autoencoders (VAE) and their different varieties. With other focusses on image patch context learning, network parallelisation using cloud services, pathology slide data creation using VAE latent space, transfer learning within latent space and focussing neural networks. Another possible area of research is segmentation, classification and anomaly detection using VAE latent space, by using outlier detection and manipulation techniques. The research will also aim to create documentation on the current state of the research area, including taxonomies of VAEs and other novel techniques.
Anna-Grace Linton
Research Topic: Predicted Long Term Outcomes following Cancer from early self-evaluated Quality of Survival
Supervisors: Professor Vania Dimitrova, Professor Adam Glaser, Dr Amy Dowling, Dr Richard Wagland (University of Southampton)
About Anna: Anna was the first student representative for the UKRI CDT for Artificial Intelligence for Medical Diagnosis and Care. She graduated from University of Bristol with a Masters in Neuroscience.
Project Description: Patient Reported Outcome Measures (PROMs) are patient self-reports which provide information on perceptions and experiences of treatment through Likert scales responses and free text comments. In assessing cancer survival, PROMs are an increasingly valuable tool to investigate the factors that influence the quality of survival of cancer patients. This has implications for long term outcomes such as health service utilisation and mortality. My PhD project will apply Artificial Intelligence (AI) methods to PROMs datasets to produce predictive models for long term outcomes following cancer treatment. This will be enriched by automated analysis of the patients’ unstructured free text comments using natural language processing (NLP). The project will further the understanding of the factors that are key predictors of outcomes of cancer patients and, in turn, improve survival and their quality of life following treatment.
Sam Llanwarne
Research Topic: Using AI to combine genomic with pathology Images to improve Lymphoma Diagnosis
Supervisors: Dr Sharon Barrance, Dr Cathy Burton, Professor David Westhead
About Sam: Sam studied Physics and Astronomy at Durham University followed by a Masters in AI and Computer Science at St Andrews University.
Project Description: In my project I will be looking at lymphoma image and genetic data, with the intention to contribute to the understanding of different sub-types of lymphoma and to improve diagnosis. Lymphoma arises from excessive mutations in lymphocytes. The pathogenic variants often occur during lymphocyte differentiation, where induced variation is crucial. Currently, researchers have defined over 60 sub-types of lymphoma, although these are usually grouped into broader categories. In this project I hope to utilise a range of tools based on deep learning and supervised and unsupervised techniques to identify lymphoma sub-groups and better understand the disease. For this, over 1500 data samples have been made available. Each data sample has genetic data and image data, where each image of the tissue sample is high resolution, featuring microscopic detail, resolving features down to cell nuclei.
Joe Sims
Research Topic: Spatial Data Mining in Histopathology
Supervisors: Dr Heike Grabsch, Dr Derek Magee
About Joe: Before joining the CDT, Joe graduated with a masters in Astrophysics from Liverpool.
Project description If you are diagnosed with stage II/stage IIIb gastric cancer and live Japan, you will most certainly be treated with surgery followed by adjuvant chemotherapy. In many countries, it is common practice to undergo this treatment without question, but what if a subset of patients were able to recover from surgery alone without the addition of chemotherapy? As part of the CLASSIC Trial, hundreds of tissue samples containing tumour were extracted from patients with stage II/stage IIIb gastric cancer, half of which received surgery and the other half received surgery and chemotherapy. Using a combination of deep learning and spatial analytical techniques, we aim to identify the patients who won’t benefit from the addition of chemotherapy and provide a pathological explanation as a result.

2020 Cohort

Research Topic: Developing Artificial Intelligence Solutions in Ovarian Cancer Diagnosis
Supervisors: Dr Nishant Ravikumar, Dr Nic Orsi, Dr Kieran Zucker, Prof Geoff Hall
About Jack: I have a maths integrated masters and a year of experience as a data analyst - essentially, I already know about Machine Learning and I'm beginning to turn that knowledge towards medical research.
Project description Ovarian cancer (OC) carries the heaviest mortality burden of all gynaecological malignancies, with 7,400 new diagnoses per year in the UK alone. Its diagnosis has been subject to the digital revolution that has taken place in histopathology over the last decade, with clinical departments progressively moving towards working with digital images of histology slides. Whilst this technological advance offers histopathologists advantages in terms of smooth workflow, remote access and low volume storage, the corollary is the generation of petabytes of digital information. The latter has been the focus of a number of endeavours in artificial intelligence (AI) aimed at extracting diagnostic information from digital images, with promising early results in defining grade and molecular profiles across a range of malignancies. The project aims to extend these pioneering approaches to OC by (1) differentiating between ovarian lesions in the progression to invasive carcinoma, (2) diagnosing across the spectrum of ovarian tumours, (3) classifying tumour grade (e.g. low vs. high grade in serous lesions), (4) identifying tumour molecular profiles, and (5) developing models of markers of response to therapy.
Research Topic: AI Cancer RATS in Primary Care
Supervisors: Mr Owen Johnson, Dr Marc de Kamps, Dr Ciarán McInerney, Prof Richard Neal
About Emma: My background is in Maths and Computer Science (BSc) and I’ve just finished my MSc in Health Informatics, my thesis was centred around AI fairness in healthcare, but I am also interested in multiple aspects of medical AI.
Project description Cancer RATS are Risk Assessment Tools that implement a simple set of rules to help primary care doctors make an early assessment of the risk of cancer. The project is to develop AI based approaches that will help improve this approach to the early detection of cancer risk. It will be based on a retrospective cohort study of data from ResearchOne, assessing the positive predictive value of Cancer RATS over a historic moving time frame to better understand the key data points which trigger an increase in patient risk and to determine the scale of missed opportunities. The research will use this data to develop next generation Machine-Learning/Process Mining algorithms designed to improve upon the current generation of Cancer RATS.
Research Topic: Patient Safety with AI – using AI to ensure safer AI
Supervisors: Dr Marc de Kamps, Mr Owen Johnson, Dr Ciarán McInerney, Dr Tom Lawton, Dr Alwyn Kotze, Dr Ibrahim Habli (University of York)
About Siân: My background is in maths and I have just graduated from an integrated masters in maths at the University of Leeds.
Project description Patient safety is a well-established discipline, encompassing the avoidance, prevention and amelioration of adverse outcomes or injuries stemming from the process of healthcare. The patient safety of AI solutions is a growing concern, particularly where the design and development of AI does not consider potential hazards of its use, misuse and abuse. This research will apply methods and heuristics, borrowing from statistical theory, information theory and adversarial machine learning, to develop an AI solution that can assess other AI solutions in cancer care, with regards to the patient safety risks arising from bias in the solutions. The project has the support of the Yorkshire and the Humber NIHR Patient Safety translational Research Centre.
Research Topic: Detection of recurrence in cancer patients
Supervisors: Mr Owen Johnson, Prof Geoff Hall
About Alex: MSc from the University of Nottingham in Physics with Astronomy with a Machine Learning focussed Masters project.
Project description The research will build on work commenced within a Macmillan funded programme linking primary care records to hospital records from the Leeds Cancer Centre. The program will apply advanced analytics to the clinical record including both structured and unstructured events to identify cancer recurrence events which have been miscoded or missed within the structured hospital record. The research will build on an analysis of structured data within the health record (chemotherapy, radiotherapy, surgical events), and natural language processing of plain text reports from radiology and other clinical areas.
Research Topic: Temporal Graph-based Convolutional Neural Networks for Electronic Healthcare Record Data
Supervisors: Dr Sam Relton, Dr Andrew Clegg, Prof Philip Cohaghan, Dr Sarah Kingsbury
About Zoe: I have just completed a Masters in Medical Engineering, and I look forward to getting to learn more about behind the scenes of in the NHS.
Project description This project is concerned with the development of a novel neural network architecture for early prediction of disease from electronic heath records. The key idea is to use a graph-based approach to encode the temporal occurrence of clinical events, coupled with graph convolutional methods to perform inference. The work will begin with a preliminary study on musculoskeletal disease and frailty, using the existing ResearchOne dataset. The application to cancer prediction will follow this proof of principle study, alongside exploration of the impact of multimorbidity on the patient trajectory.
Research Topic: Predicting cancer outcomes from patient reported data and routine healthcare data
Supervisors: Prof Tony Cohn, Prof Vania Dimitrova, Prof Adam Glaser, Dr Amy Downing
About Shazeea: I have an integrated masters in Biochemistry and since graduating, have been working for a Pharmaceutical company called Novartis in Medical Affairs/ project management.
Project description Cancer survival in the UK has doubled in the last 40 years; 50% survive cancer for 10 or more years (2010-11 data). However, UK survival rates are still below the rates in similar countries. Predicting long term outcomes following cancer treatment is crucial. A Patient Reported Outcome Measure (PROM) is a report coming directly from patients about how they feel or function in relation to a health condition and its therapy. Cancer patients’ PROMs represent an important part of assessing patient quality of life, indicating whether treatment has improved a patient’s symptoms, the type of experience of care patients have received at the practice, whether the patient’s health and well-being is improving. The aim of the project is to determine whether AI techniques can enable accurate prediction of cancer outcomes using PROMs & routinely collected health data to inform the production of robust personalised outcome prediction models addressing quality of survival as well as absolute duration of survival.
Research Topic: Application of NLP to automated COSD in Ovarian cancer and other malignancies
Supervisors: Dr Serge Sharoff, Prof Geoff Hall
About Sarah: My undergraduate degree was in Information Systems and  my MSc was in Data Science. Prior to the CDT starting I was working at Jet2.com as a Data Engineer.
Project description The work proposed will apply NLP techniques to the automated extraction of structured data required for the completion of the COSD dataset from clinical, radiological, pathological and surgical records. The tool would guide the clinician to include critical features in the annotations and reports defined in a disease setting relevant manner providing interactive feedback in real-time to critical fields for which structured data has not been included or is unclear.
Research Topic: AI analysis of voice to aid laryngeal cancer diagnosis
Supervisors: Dr Luisa Cutillo, Dr James Moor
About Mary: I graduated from the University of Birmingham with an integrated masters in electronic and Electrical Engineering.
Project description Patients with head and neck cancer present to secondary care services in a number of ways. Laryngeal carcinoma, (cancer affecting the voice box) tends to cause hoarseness when it affects the vocal cords. There is an urgent clinical need to develop tools to aid in the diagnostic process for patients referred on the NHS ’2 week wait’ pathway. The research will focus on assessment of a patient’s voice to aid clinical decision making as to whether the larynx is a site of potential cancer development, or, whether it is affected by a wide range of non-cancer diagnoses. 
Research Topic: Coverage measure and navigation strategies for robotic colonoscopy: Fusing machine learning and robotics to enhance diagnosis and treatment
Supervisors: Prof Pietro Valdastri, Dr Venkat Subramanian
About Sarah: I have a BSc and MChem in Chemistry from Leeds and currently finishing up my MSc in Computer Science from Nottingham. My project was on using support vector machines for Alzheimer’s disease classification. My current research interests are mainly in medical diagnosis using biomedical imaging and treatment.
Project description Over the last decade, the STORM Lab (https://www.stormlabuk.com/) has developed the Magnetic Flexible Endoscope (MFE), an innovative platform for easy and painless colonoscopy. In conventional colonoscopy, 1 in 20 polyps are not detected during visual exploration due to the unstructured environment (e.g., tissue folds). Moreover, the navigation is difficult due to tight turns and obstacles. The project will explore the use of machine learning to integrate the localisation information provided by the MFE platform with the visual information produced by the endoscopic camera, in order to (1) estimate the total percentage of the colon visualised and (2) generate a navigation strategy to increase the precision of locomotion and endoscopic tasks such as biopsy.
Research Topic: Transfer learning and transferrable models in Digital Pathology for Gastrointestinal Cancers
Supervisors: Dr Derek Magee, Prof Heike Grabsch, Dr Nicholas P West
About Wilson: I have a background in Computer Science and have been working at the Leeds Institute for Data Analytics as a intern and research assistant since graduating in 2018. My previous work contributed to the development of novel simulation software’s for population projection and forecasting, network analysis with unsupervised machine learning in a comparative study of bike share schemes as well as writing my BSc dissertation on the application of Machine Learning and Data Science for analytics and diagnostics.
Project description Digital Histopathology is the use of whole slide images of scanned stained tissue sections to diagnose and study diseases such as cancer. One issue with histologic imaging data is that tissue samples and acquired images may vary due to different methods of sample within the same hospital or between different hospitals. Additionally, samples from different parts of the body, different parts of the world, or from patients of different ages can have different appearance. The project aims to investigate and establish methods for automatically adapting models to new data sets including the way in which such unsupervised model transfer/learning could be automatically or semi-automatically validated. The research will have access to a dataset on 5000 patients with oesophagogastric cancer from over 150 hospitals in four countries around the world.