Aksoy Laboratory
The Aksoy Laboratory develops AI systems for complex decisions in healthcare, research and human performance, and studies how they can be translated responsibly into real-world use.
Our research is organised around the full AI lifecycle. We work across four stages: build, validate, deploy and govern, and study what knowledge transfers between them. This includes how real-world needs shape model design, how performance and uncertainty generalise across populations and settings, how tools integrate with workflows, and how safety, equity and accountability are maintained over time.
Our clinical programmes span oncology, dermatology and ophthalmology, including cancer prognosis, melanoma risk and screening, retinal imaging and childhood myopia. We also collaborate on AI for health-service operations, including pressure-injury risk and patient-flow prediction; high-performance sport, including cricket; and research intelligence through the Research Activity Monitoring System (RAMS), which generates AI-powered insights for NSW Health. These applications provide test beds for methods in multimodal prediction, external validation, explainability, privacy-preserving analysis and responsible implementation.
We design for translation from the outset, integrating real-world value, technical performance, governance, regulatory readiness and pathways to adoption or commercialisation. We welcome partnerships with patients and consumers, clinicians, researchers, health services, government, sporting organisations, industry and investment partners who bring an important decision problem, distinctive data, implementation capability or a pathway to impact.
Biomedical AI
- Clinical artificial intelligence
- Dermatology, melanoma and skin cancer
- Cancer risk prediction and prognosis
- Screening, early detection and health policy
- Ophthalmology, retinal imaging and oculomics
- Paediatric myopia
- Ultraviolet radiation and disease of the skin and eye
- Federated learning, data privacy and sovereign AI
- Population-scale data linkage and health intelligence
- AI governance, ethics and regulation
- Clinical prediction modelling: nomograms, ensemble machine learning, external validation, calibration, decision-curve analysis
- Explainable AI: SHAP attribution and Bayesian uncertainty quantification
- Dermatological imaging AI: 3D total body photography, dermoscopy, standardised photodamage grading
- Ophthalmic imaging AI: retinal foundation models, fundus and OCT analysis, ocular biometry
- Decision-analytic methods for screening policy: net benefit, overdiagnosis adjustment, capacity-constrained threshold setting
- Generative AI for medical imaging: latent diffusion models, synthetic image validation
- Federated and privacy-preserving learning, including differential privacy budget allocation
- Population-scale linked data analysis (cancer registry, Medicare, pharmaceutical and Census data; ABS DataLab)
- Multi-omic biomarker analysis: metabolomics, proteomics, lipidomics, genetic risk scoring
- Deployment engineering: turning a published model into a maintained clinical tool
- Regulatory pathways for clinical decision support software
- Novel prognostic nomogram for predicting recurrence-free survival in medullary thyroid carcinoma. Histopathology, 2024. pubmed.ncbi.nlm.nih.gov/38253940/
- A novel prognostic nomogram for predicting survival in diffuse pleural mesothelioma. Pathology, 2023. pubmed.ncbi.nlm.nih.gov/36842876/
- Development of a novel malignant colorectal polyp prognostic nomogram. ANZ Journal of Surgery, 2025. pubmed.ncbi.nlm.nih.gov/40616247/
- Spatial and temporal control of CRISPR-Cas9-mediated gene editing delivered via a light-triggered liposome system. ACS Applied Materials and Interfaces, 2020. pubmed.ncbi.nlm.nih.gov/33174413/
- Novel prognostic nomogram for predicting recurrence-free survival in medullary thyroid carcinoma. Histopathology, 2024. pubmed.ncbi.nlm.nih.gov/38253940/
- A novel prognostic nomogram for predicting survival in diffuse pleural mesothelioma. Pathology, 2023. pubmed.ncbi.nlm.nih.gov/36842876/
- Development of a novel malignant colorectal polyp prognostic nomogram. ANZ Journal of Surgery, 2025. pubmed.ncbi.nlm.nih.gov/40616247/
- Spatial and temporal control of CRISPR-Cas9-mediated gene editing delivered via a light-triggered liposome system. ACS Applied Materials and Interfaces, 2020. pubmed.ncbi.nlm.nih.gov/33174413/
Student Opportunities
Predicting melanoma risk, and deciding who to screen
Projects: Honours, Masters & PhD
Project details
Primary Supervisor: Dr Yagiz Aksoy
Supervisory Team: Associate Professor Daniel Hesselson
Centre: Biomedical AI
Australia has the highest melanoma incidence in the world, yet risk is still assessed with short questionnaires, and no one can yet say at what risk score screening should begin. This project spans both halves of that problem. Students build and validate melanoma risk calculators using ACEMID, a national collaboration that has captured whole-body 3D photographs of thousands of Australians, testing whether measurable image phenotypes — naevus counts, freckling, ultraviolet damage, objective skin colour — outperform the questions currently asked. Those calculators then feed adaptive screening thresholds: decision-analytic models that weigh clinical benefit, overdiagnosis and real workforce capacity, region by region, for a national screening programme under active design. The work sits inside national collaborations with a direct route into policy. Suits students with a computational, statistical, epidemiological or engineering background; clinical knowledge welcome but not required.
Oculomics: the eye as a window to the body
Projects: Honours, Masters & PhD
Project details
Primary Supervisor: Dr Yagiz Aksoy
Supervisory Team: Associate Professor Daniel Hesselson
Centre: Biomedical AI
A single retinal photograph carries signal about the eye and about the body behind it. Oculomics is the emerging discipline of reading that signal at scale, and this laboratory sits on one of the strongest platforms in the country to pursue it: more than 300,000 retinal scans within the Global RETFound consortium, and a macular degeneration biobank linking blood biomarkers — plasma proteomics across 9,500 proteins, targeted lipidomics, combined nuclear and mitochondrial genetic risk scores — to imaging. Students build AI models that connect what the eye shows to what the blood measures, asking how much earlier eye disease can be flagged and how far systemic signals can responsibly be read from the retina. Suits students with a machine learning, biostatistics, bioinformatics or vision-science background.
AI predictive tools for childhood myopia and glaucoma
Projects: Honours, Masters & PhD
Project details
Primary Supervisor: Dr Yagiz Aksoy
Supervisory Team: Associate Professor Daniel Hesselson
Centre: Biomedical AI
One in three Australian children is now myopic, and early onset carries lifelong risk of retinal detachment, glaucoma and myopic maculopathy — yet treatment options have multiplied faster than the evidence to choose between them. Students design explainable predictive tools: models that forecast how fast a child’s eye will grow, translate that trajectory into lifetime risk of high myopia and its complications including glaucoma, and recommend when to escalate treatment — built on an eight-year Australian paediatric cohort, and designed so every recommendation can be explained to the clinician making the call and the parent hearing it. Suits students with a machine learning, engineering or optometry/orthoptics background
Additional Opportunities:
Your idea
Projects: Honours, Masters & PhD
Some of the best projects in this laboratory will be ones we have not thought of yet. We welcome talented students who arrive with their own question — a dataset you want to interrogate, a clinical problem you believe AI can genuinely change, a method you want to test against the real world. The laboratory’s pipeline from model to deployed, maintained tool is available to a good idea regardless of where it starts. Send us a one-page sketch and we will take it seriously.