AI for Endoscopy Video Analysis
AI-based computer-aided diagnosis systems for endoscopy video applications are an area of rapid research and development. One application is the real-time computer-aided detection and diagnosis of colorectal polyps during colonoscopy, a crucial examination for the screening and prevention of colorectal cancer (currently the fourth most diagnosed and the third most fatal cancer worldwide).
The School of Science and Technology at City, University of London has partnered with Cosmo IMD (world-leading company in the design of AI tools for endoscopy) to offer a Doctoral Scholarship to investigate novel approaches for endoscopy video analysis. The Scholarship will focus on self-supervised methods and on their power to harness unlabelled data for downstream tasks.
Closing date: 15th of Jun 2023 or until the position has been filled.
The successful candidate will work in an exciting international environment in the heart of the City of London. They will join the CitAI Research Centre (which features academic staff with extensive expertise in machine learning for healthcare) and be able to exploit the power of Hyperion, City’s new High-Performance Computer. They will also have access to part of Cosmo IMD’s proprietary library of annotated endoscopy videos (of unprecedented size, quality, and heterogeneity) and be supported by their expertise in industrial deployment of cutting-edge, FDA-approved AI tools for colonoscopy. The research outputs of this Scholarship will have the chance of impacting the real clinical practice.
Project outline
One of the downsides of deep learning algorithms is that they are significantly data-hungry, requiring vast amounts of manually labeled samples to function correctly and accurate annotations to achieve good performances. In most healthcare applications, including endoscopy video analysis, obtaining such video data is costly (due to the need of medical expertise for accurate annotation) and time-consuming (due to the video length of endoscopy applications often lasting hours). Furthermore, many endoscopy applications focus on the detection and classification of clinical conditions or lesions located in a small fraction of video frames or involve significant class imbalances.
Self-supervised learning refers to methods that use unlabeled data on a pretext task to learn deep models able to capture rich visual data representations without explicit supervision. These pre-trained models are then finetuned with limited labeled data on a downstream task of interest. Self-supervised learning approaches proved to be capable of achieving remarkable results in many computer vision applications, matching or even surpassing the performance of fully supervised models, showing promise to become the new gold-standard pre-training approach for deep learning models. In the same fashion, we envision that successful self-supervised learning on unlabeled endoscopy videos will constitute in the future the building block for the development of high-capacity deep learning models that are robust and accurate on different downstream endoscopy tasks.
What is offered
The doctoral studentship will include:
- An exceptionally competitive annual bursary for 3 years (£24,000/year)
- Full tuition fees for UK Students (and partial coverage for EU and International ones).
- The opportunity to earn up to around £4,300/year through a non-compulsory teaching assistantship.
- Over £7000 to participate to conferences and training.
Eligibility and requirements
The studentships will be awarded based on outstanding academic achievement and the potential to produce cutting-edge research. Prospective applicants must:
- Hold a good honours degree (normally no less than a second-class honours degree or an equivalent qualification) in an appropriate subject (including computer science, mathematics, engineering, physics). Exceptionally, if the first degree is in a different subject area, we can consider applications from those with a good Master’s Degree in a relevant subject or extensive professional experience in the area of their proposed research.
- Applicants whose mother tongue is not English must meet any one or a combination of the following:
→ A minimum IELTS average score of 6.5; with a minimum of 6.0 in each of the four components.
→ The award of a Masters’ degree, the teaching of which was in English from an English-Speaking Country.
How to Apply
To apply online, click on this link.
Applicants are welcome to discuss their application in advance with the academic supervisor in the Department Dr. Giacomo Tarroni, giacomo.tarroni@city.ac.uk.
The successful candidate will formally start their doctorate either in July or in October 2023.
The PhD application form requires candidates to attach a research proposal. Applicants to this studentship do not need to do so. Instead, they should enter the title of the research project: AI for Endoscopy Video Analysis and supervisor Dr. Giacomo Tarroni. The entry point will be in October or later.