Medical AI PhD: Screening for Diabetic Retinopathy with Fundus Imaging and Smart Cameras

Medical AI PhD: Screening for Diabetic Retinopathy with Fundus Imaging and Smart Cameras

Quick Decision

The stated core is automatic 0–4 grading of diabetic retinopathy in fundus images, with interpretable outputs and comparison against NHS clinician assessments. The brief also lists glaucoma, cataract, age-related macular degeneration, and OCTA/CCM work on diabetic neuropathy, leaving the primary disease and modality scope broad. It suits applicants interested in clinical image analysis and AI; before applying, clarify the 500,000-image mix, dataset approvals, and whether “smart cameras” includes hardware development. Official project PDF DLA award page

Research Problem and Doctoral Work

What Is Unsolved and Why It Matters

The clinical motivation is access to diabetic eye screening, but the advertised background uses an overgeneralized annual-screening claim. Current England guidance invites eligible people aged 12 and over onto 12-month or 24-month pathways according to screening history and outcome. A current UK National Screening Committee evidence review says automated retinal grading is not yet recommended for routine incorporation into the diabetic eye-screening programme. Some systems perform similarly to first human graders for more severe disease, but disease-free detection, site/vendor differences, possible age or ethnicity effects, patient and service outcomes, and UK cost-effectiveness remain uncertain. NHS England screening pathway standards · UK NSC 2026 evidence review

That context makes the strongest research question more demanding than “can a model classify fundus images?” A useful study would define the clinical threshold and reference standard, keep patient-level data splits, separate sites or cameras for external testing, report sensitivity and specificity at a justified referral threshold, examine calibration and ungradable images, and assess subgroup performance. Explainability should be tested with intended users and clinical workflow rather than judged only from visual saliency maps. These are evidence-led research angles, not promised work packages. UK NSC external evidence review

Work, Methods, and Potential Outputs

The official project document proposes developing and evaluating AI, machine-learning, deep-learning and data-science methods for medical image analysis. Official project PDF The concrete advertised work includes:

• automatically grading retinal fundus images into diabetic-retinopathy severity levels 0–4;
• optimizing model explanations so clinicians and patients can understand the diagnostic basis;
• comparing algorithm grading with NHS clinician grading;
• exploring glaucoma, cataract and age-related macular degeneration, alongside diabetic retinopathy;
• exploring OCT angiography (OCTA) and corneal confocal microscopy (CCM) for diabetic neuropathy;
• using an intended total of about 500,000 fundus images from multiple public and customized datasets.

No architecture, work-package timeline, reference standard, case-level split, primary disease, or specific camera build is stated. The award listing uses “Smart Cameras,” while the detailed project PDF describes image analysis and does not specify a hardware design or prototype task. Confirm what that label means.

The data list also needs a project-level inventory before feasibility can be judged. EyePACS describes more than five million images but asks researchers to contact it about access and use. EyePACS research access The UK Biobank count of 175,824 fundus images from 85,848 participants is a historical October 2019 snapshot; the current UK Biobank imaging page describes a different retina modality, OCT, available for 100,000 participants. New UK Biobank applications are currently paused during Research Analysis Platform changes, with the organization intending to accept new applications in late 2026. Historical UK Biobank fundus-image paper · Current UK Biobank imaging data · Access application status

INSIGHT's researcher pages describe a formal application and secure research access. Its April 2026 dashboard lists 31,288,068 eye images, whereas the project PDF says more than 35 million. The figures may represent different snapshots or definitions; ask which release and modalities are planned. INSIGHT researcher access · INSIGHT April 2026 dashboard HVDROPDB is a more substantial scope issue: its primary data paper describes retinal images of premature infants for retinopathy of prematurity, not diabetic retinopathy. Ask whether it is intended for a separate ROP task or was included by mistake. HVDROPDB primary dataset paper North Devon and Royal Devon NHS data and any India rural-image collaboration also require confirmation of consent, governance, labels, cross-site access and storage.

The PDF ends with a claimed benefit that automation will increase the chance of irreversible damage occurring before retinopathy is detected. That contradicts its stated early-detection and reduced-undetected-cases rationale and is likely a wording error; ask the supervisor to clarify.

School, Programme and Field Context

The degree-awarding institution is the University of Exeter, with the proposed programme listed as PhD Clinical and Biomedical Sciences. The project PDF gives the work location as the RILD building in Exeter. RILD sits at the Royal Devon hospital site and is a university/NHS-associated research and education facility. The project names an NHS ophthalmologist as a co-supervisor, but the material does not describe a separate employer, joint degree, or automatic clinical-data entitlement.

The Clinical and Biomedical Sciences research environment explicitly includes data science and AI, diabetes, epidemiology and public health, experimental medicine and precision medicine, and describes close working with the NHS and NIHR. The Doctoral College offers general researcher training and professional development; a project-specific technical curriculum and attendance schedule are not published.

Relevant subject signals are stronger than a generic whole-university ranking: Exeter's REF 2021 Clinical Medicine unit reported 45% 4-star and 48% 3-star research. The programme page reports THE World University Rankings 2026 positions of Top 125 Life Sciences and Top 175 Medical & Health. These are institutional subject signals, not a measure of this individual project or supervisor team.

Exeter's full-time PhD registration maximum is four years, while this award funds up to 3.5 years. Applicants should ask how registration, training, thesis submission and any period after funding ends are arranged. The specific studentship instructions override the general programme page's self-funded application route.

Supervisor and Lab Assessment

Professor Neil Vaughan is named as lead supervisor. The PDF names Dr Conor Ramsden, an NHS ophthalmologist, and Professor Xujiong Ye as co-supervisors. Exeter's profile places Vaughan in Clinical and Biomedical Sciences at RILD. Ye's ORCID lists a current Exeter professorship in computer vision and AI in healthcare alongside a continuing Lincoln professorship; confirm the practical supervision split and availability directly. The project mentions an established medical-image-analysis lab, but does not give its composition, compute resources, or day-to-day mentorship arrangement. Neil Vaughan profile · Xujiong Ye ORCID profile

Three recent co-authored papers help interpret the team's relevant research:

• Vaughan's 2026 scoping review examined 30 smartphone fundus hardware solutions and 274 sources. It identifies potential for screening or triage, but reports variation in performance across disease, device, protocol and operator, plus unresolved governance, safety, regulation and cost-effectiveness issues. This supports asking what “Smart Cameras” means; it does not prove that this PhD includes hardware work. Smartphone fundus-imaging review
• A 2025 MICCAI paper co-authored by Ye uses ordinal regression and diffusion-conditioned probabilities to model diabetic-retinopathy grades, evaluated on four public fundus datasets. Because grades 0–4 have a natural order, this is relevant methodological context, not a promised architecture or proof of clinical validity. MICCAI 2025 paper
• A 2026 IEEE Transactions on Medical Imaging paper co-authored by Ye addresses long-tailed grade distributions and ambiguous severity thresholds through autoregressive ordinal decisions and geometric priors. It reports results across 2D and 3D medical grading datasets, not necessarily the datasets or outcomes proposed here. IEEE TMI 2026 paper

A 2026 England target-product-profile consensus also illustrates the implementation bar: stakeholders agreed specifications for a diabetic-eye-screening AI product spanning clinical validity and utility. It is a system-requirements paper, not evidence that a particular algorithm is safe, effective or commissioned. England target-product-profile consensus

Funding, Application, and Living Cost

Item	Verified terms and caveats
Award	EPSRC Doctoral Landscape Award, project-led and competitive; up to 16 Exeter studentships for September 2027 entry
Funding	Up to 3.5 full-time years; tuition fees and a £5,000 Research Training Support Grant over 3.5 years
Stipend	£21,805 per year is listed and index-linked, but pages conflict on whether this is the 2026/27 current rate or 2027/28 rate; confirm the actual 2027/28 amount
International fees	Exeter says it covers the Home/International fee difference for successful international DLA recipients; international fee-status places are capped at up to 30% of the cohort
Other international costs	Student visa, Immigration Health Surcharge and moving costs are the student's responsibility
Deadline	Midday, 8 December 2026; applications must use the project link in the official PDF
Documents	Project-named personal statement, CV, transcripts, two references, and English-language evidence if applicable; documents must be in English, with certified translations of qualifications
Academic rule	Normally a First or Upper Second in an appropriate science, engineering or technology subject or equivalent; a Lower Second may be considered with a master's or significant relevant non-academic experience
Selection limit	Applicants may submit up to three separate project applications and should name a preference if applying to more than one

The project award record sets references due by 8 January 2027 and general interviews likely 15–24 February 2027. The central EPSRC guidance instead says references are due 15 February 2027 and still includes a 2026 interview/outcome calendar despite describing 2027 entry. Ask Exeter admissions to confirm the schedule; plan for references to arrive by the earlier 8 January date. Project award schedule · Central EPSRC DLA schedule

Exeter lists 2026/27 research tuition at £5,238 for Home and £28,900 for International students; its 2027/28 page shows £29,800 International fees and Home fees still TBA. These figures provide context only: successful DLA award holders receive tuition coverage under the award terms. EPSRC DLA award details

The university's April 2026 estimate for a single student in Exeter is £1,300–£2,000 per month. Dividing the currently listed stipend by 12 gives about £1,817 per month: an indicative margin of roughly £517 at the lowest budget estimate and a shortfall of about £183 at the upper estimate. Rent choice and one-off costs matter. This comparison excludes visa, health surcharge, travel, relocation and emergency savings. Exeter living-cost estimate The stipend-year conflict should be resolved before accepting.

Applicant Fit and Preparation

Required Background and Skills

The general DLA academic rule is an appropriate science, engineering or technology degree at First/2:1 level or equivalent; Lower Second applicants may be considered with a master's or significant relevant non-academic experience. The programme page gives further general background expectations. The project itself does not list a separate mandatory technical skill.

Preferred Experience and Suggested Preparation

The project PDF explicitly says the following are beneficial but not essential: motivation to contribute to medical technology, software development or coding, Python machine learning/data science, and medical image analysis/image processing.

A useful application portfolio would show a reproducible Python image-analysis project and explain patient-level splits, class imbalance, validation limits, image quality, and the difference between benchmark performance and clinical benefit. A personal statement should connect prior research, the chosen project, and motivation for working with clinical imaging. These are preparation suggestions, not extra eligibility rules.

Career Paths and Market Context

The Clinical and Biomedical Sciences programme lists broad graduate destinations including academic research, biotechnology and commercial R&D, NHS research and innovation, teaching, the civil service, charities and policy. Skills developed through this project could transfer to Python and image-processing pipelines, deep-learning evaluation and data management, clinical collaboration and evidence translation, explainability and bias assessment, responsible AI, and medical-device or NHS evidence work. Exeter’s programme page describes these sectors; the DSIT AI Labour Market Survey 2025, published in January 2026, describes UK-wide AI role trends, skill gaps and changing needs. It is not a vacancy forecast for ophthalmic-imaging jobs. The project advertises no placement rate or employer commitment. Clinical AI roles may require stronger software engineering, regulatory knowledge, prospective validation or clinical-domain experience beyond a PhD model benchmark. (Clinical and Biomedical Sciences PhD programme, DSIT AI Labour Market Survey 2025)

Next Actions and Questions to Verify

Before Applying

1. Ask which condition and modality define the thesis's central question and what will be removed from the broad project list.
2. Request the exact 500,000-image inventory, planned dataset-by-task mapping, access status, labels, permissions, approvals and secure-computing arrangements.
3. Ask whether HVDROPDB is intended for a separate retinopathy-of-prematurity task and which INSIGHT release is planned.
4. Clarify whether “Smart Cameras” means hardware design/prototyping or image analysis only.
5. Confirm that the applicant meets DLA residence and degree requirements and that project-specific eligibility adds no other condition.
6. Ask Exeter admissions to resolve the 8 January versus 15 February reference-date conflict and confirm the 2027 interview/outcome schedule; until then, prepare references by 8 January. Use the application link inside the official project PDF.

At Interview

1. Ask for the primary endpoint, NHS grader roles, adjudication/reference standard and handling of ungradable images.
2. Ask whether evaluation will use patient-level and external site/device splits, calibration, and age/ethnicity subgroup analyses.
3. Ask how explanations will be evaluated with clinicians or patients, and who will supervise the clinical and computational parts day to day.
4. Ask how the project will be narrowed across diseases, modalities, datasets and any camera work.

Before Accepting an Offer

1. Obtain written confirmation of the 2027/28 stipend, total award length, tuition coverage, £5,000 RTSG terms and any research costs not covered.
2. Confirm how the 3.5 funded years align with the four-year maximum PhD registration and thesis submission date.
3. For an international award, confirm fee-waiver application and cohort eligibility, then budget visa, IHS and relocation separately.
4. Ask what data approvals are already in place and whether any access depends on future applications.

Evidence and Sources

Project, programme and university

• EPSRC DLA project PDF
• EPSRC DLA award record, reference 6010
• EPSRC DLA eligibility and application information
• Clinical and Biomedical Sciences PhD programme
• Exeter REF 2021 results
• Exeter Doctoral College
• Exeter PhD degree durations
• Neil Vaughan's Exeter profile
• Xujiong Ye's ORCID profile

Recent research and clinical evidence

• Smartphone-based fundus imaging in global eye care (2026)
• MICCAI 2025 paper on ordinal diabetic-retinopathy grading
• IEEE TMI 2026 DiffGeo-AOR grading paper
• England diabetic-eye-screening AI target product profile consensus
• UK NSC consultation on AI in diabetic eye screening
• UK NSC 2026 external evidence review
• NHS England diabetic eye screening standards
• Diabetic eye screening pathway requirements
• MHRA software and AI as a medical device guidance
• NICE Evidence Standards Framework for digital health technologies

Dataset access and cost

• EyePACS research data access
• UK Biobank fundus-imaging study and 2019 image count
• UK Biobank application and access status
• UK Biobank current imaging data, including OCT
• INSIGHT researcher access overview
• INSIGHT April 2026 data dashboard
• HVDROPDB primary dataset paper
• Exeter international student budget estimate
• DSIT AI Labour Market Survey 2025