Verjans Liao Lab · AI Methods: Biology + Imaging · AIML, Adelaide University

Foundational machine learning, taken into a working health system.

We build multimodal AI across images, text, sensors and omics, with a focus on explainability, generalisability and causality, and on the small and noisy data that clinical work actually produces. We publish at NeurIPS, ICML, CVPR and ICCV, and we have taken three AI systems into SA Health.

Our research AI deployed in the clinic

Members of the Verjans Liao Lab together outdoors in the Adelaide Hills, with the city and coastline behind them.

Partners SA HealthSAHMRIRoyal Adelaide Hospital MedtronicSiemens HealthineersSanofi Heart FoundationUMC GroningenFlinders

Figures last checked September 2026. Combined citation and publication counts include co-authored work in each profile.

What we work on

Five method themes, one clinical test

AI deployed in the clinic

How they got there →

Latest updates and news

All news →

Research

Research themes

The group is co-led by a practising cardiologist and a computer vision researcher. Our methods work runs along five themes: multimodal AI, explainability, generalisability, small and noisy data, and causality. Each is tested against the same question, which is whether a clinician would act on the output.

Venue standing

Where the work is published

By Google Scholar Metrics 2025 h5-index, CVPR ranks #1 in computer vision and #2 among all publications worldwide; NeurIPS is #1 in artificial intelligence; ECCV #2 and ICCV #3 in computer vision; ICML #3 in AI; MICCAI #4 in radiology and medical imaging. In 2025 the group had 2 papers accepted at CVPR, 3 at ICCV and 1 at NeurIPS, and took the DICTA Best Paper; in 2026, 2 more CVPR acceptances, with 6 papers under submission at NeurIPS. Since 2022 that totals 5 CVPR, 7 MICCAI, 2 ICCV and one each at NeurIPS, ICML, ECCV and IJCAI, alongside work in npj Digital Medicine and three papers in JACC: Advances.

Translation

Deployment

Deploying into a working health service requires clinical governance, data sovereignty, safety evaluation and clinician adoption to be worked out alongside the model. Three have been through it in South Australia, with a fourth in development.

Systems in the health service

What deployment actually cost

RAPIDx AI established the clinical-governance and integration pathway that the two current systems are built on. AutoMedic's funded business case projects roughly 58,500 pharmacist hours freed a year, worth more than $3M, and more than 1,300 hospital readmissions avoided.

Earlier translation

The first intracoronary near-infrared fluorescence molecular imaging of atherosclerosis in a porcine model (JACC: Cardiovascular Imaging, 2016). Co-author of the PRIME checklist for evaluating cardiovascular imaging machine learning (2021), cited internationally as a standard.

Education and outreach

Clinical education

Adoption depends on clinicians being able to appraise these tools, so a large part of the work is teaching practising doctors to evaluate AI critically.

Micro-credential

Doctors + AI

“Practising Medicine in the Age of Intelligent Machines”, a new Adelaide University professional short course and micro-credential, in development with Learning Futures and the Faculty of Health and Medical Sciences.

In development

Continuing education

Clinician training

CSANZ Heart School keynote on real-world AI in cardiology (2025); ANZSGM annual scientific meeting plenary (2026); Royal Adelaide Hospital Grand Rounds (2025, 2026); education for GPs on appropriate use and referral.

Ongoing

Clinical teaching

Students and imaging staff

About 50 medical students a year on cardiology rotations at the Royal Adelaide Hospital, supervision of cardiology fellows, and more than fifty radiographers and echocardiography technicians trained in cardiac imaging since 2017.

Since 2017

Standards

Committees

Society of Cardiovascular Computed Tomography Education Committee (2026–); ESC Working Group on e-Cardiology; Publications Committee of the Society for Cardiovascular Magnetic Resonance to 2024.

Current

Invited talks since 2024

Four invited talks at ESC AI 2025, including large language models and vision-language models in cardiovascular medicine.
Keynote, Siemens Healthineers Imaging Symposium, Sydney, 2025.
Closing keynote, Spring into Clinical Trials, 2025.
Plenaries for SA Health, Cancer Australia and BioMelbourne, 2024; National Imaging Facility ASM, 2025.

Networks and editorial roles

AIML is a member of ACAIM, the Global Alliance of AI Centres in Medicine, a monthly network that includes Stanford, Mayo Clinic, Cleveland Clinic and Bern. Johan Verjans is AIML's representative in it.

Editorial board of European Heart Journal – Digital Health and Frontiers in Cardiovascular Medicine. Reviewing for JACC, the European Heart Journal and Nature journals, and for CVPR, NeurIPS, ICCV and ECCV.

Industry collaboration

Medtronic, Sanofi and the Heart Foundation are our current industry and national partners. Engagement takes the form of contract research with generated IP and a regulatory pathway, including two consecutive Medtronic contracts on ECG-based detection.

$14.1Mtotal research funding as investigator since 2022

$4.0M+held as Lead Investigator, Founder or CIA

SA Healthcollaborations across the state health service: RAPIDx, AI scribe, Pharmacy review and Cardiomics

Partners

What we offer a company

Contract research with a clinician-led group that owns the clinical question, not just the model.
A demonstrated path from algorithm to health-service deployment, including clinical governance, data sovereignty and safety evaluation.
Regulatory and deployment experience, and IP generated under two Medtronic contracts.
Access to AIML compute, with 200+ NVIDIA A100 GPUs and over 5+ PB of secure storage.

How engagements have worked

Medtronic began with a contract on heart-failure prediction from ECGs (2022–24), extended to left-ventricular dysfunction detection from external ECG (2025–26), and is now the basis of ongoing regulatory work and an international translation pathway. That sequence of contract, extension and translation is the shape we aim for.

It helps to talk early, while the clinical question is still open and before the data has been fixed.

Start a conversation

Code and data

What we have released

Our code releases live under one GitHub organisation, and our benchmark work is published openly so other groups can measure against it.

Code

github.com/AIML-MED →

Benchmarks and datasets

Public benchmark

HEAL-MedVQA

67,000 pairs

A hallucination evaluation benchmark for grounded medical multimodal LLMs: 67,000 visual-question-answering pairs with doctor-annotated anatomical segmentation masks, plus two evaluation protocols that test whether a model is reading the image or exploiting a textual shortcut. Published at IJCAI 2025 as Localizing Before Answering.

arXiv:2505.00744 · IJCAI proceedings

Clinical data

Health-service data

By collaboration

Much of our work uses South Australian health-service data: the RAPIDx AI trial cohort of about 15,000 patients across 12 hospitals, SA Pathology and SA Pharmacy records, and MyHealthRecord data under approved access. This data is governed and cannot be redistributed. Groups wanting to work with it do so through a formal collaboration, with ethics approval and a data-sharing agreement in place. That process usually sets the timeline for a project.

Reproducing our results

Each released repository carries the paper it belongs to in its description. Where a paper uses a public dataset such as MIMIC-CXR, the paediatric X-ray corpora or a public segmentation challenge, the repository documents which splits we used.

If a paper of ours has no repository yet and you need the code, please email us.

Books and standards

Artificial Intelligence in Medicine (Springer, 2022), the first comprehensive textbook on AI in medicine. Chapter on machine learning in medical imaging, senior author.

Intelligence-Based Cardiology and Cardiac Surgery (Elsevier, 2023), with two chapters, on medical visual question answering and on AI in echocardiography.

Co-author of the PRIME checklist (2021), cited internationally as a standard for evaluating cardiovascular imaging machine learning.

Publications

Selected peer-reviewed work across the lab's senior authors: Johan Verjans, Zhibin Liao and Vu Minh Hieu Phan. Author role is marked: senior = last author; supervising = final supervising-author block with technical co-leads. Filter to Joint lab papers for work co-authored by two or more of the lab's senior authors. Citation counts are Google Scholar, September 2026.

Search publications

Combined figures count co-authored work in each lead's profile. A selection, not the full record. Complete lists: Verjans on Google Scholar · Liao on Google Scholar.

People

Who is in the lab

The AIML medical machine learning group was founded in 2017 with two academics and grew past 42+ people by 2024. 10 PhD candidates are currently supervised across medicine, computer science and biomedical engineering, and 9 doctorates have completed under the leads.

Lab leads

Johan Verjans

Co-lead · Associate Professor and Future Industry Making Fellow, AIML · MD PhD FRACP FESC

Cardiologist and AI researcher. Founded the AIML medical machine learning group in 2017 with Professor Gustavo Carneiro and led it as Deputy Director (Health) from 2018 to 2024, growing it past 42 people. Platform Leader for AI at SAHMRI since 2019, and a consultant cardiologist at the Royal Adelaide Hospital and Jones Radiology. MD and PhD (cardiovascular imaging) from Maastricht University; cardiology training at UMC Utrecht; cardiac MRI fellowship at St Bartholomew's, London; postdoctoral fellow at Massachusetts General Hospital and Harvard Medical School. Level 3 (advanced) cardiac MR, EACVI and ANZCMR certified.

Sole senior author, with his own PhD student as first author, on methods at NeurIPS, ICML and CVPR; senior author in npj Digital Medicine and JACC: Advances. Founder and AI lead of AUScribe; AI lead for AutoMedic and RAPIDx AI. Member of the Heart Foundation's National Research Advisory Board (2026–) and the ESC Working Group on e-Cardiology (2025–); editorial board of European Heart Journal – Digital Health and Frontiers in Cardiovascular Medicine; AIML's representative in ACAIM, the Global Alliance of AI Centres in Medicine, alongside Stanford, Mayo Clinic, Cleveland Clinic and Bern. Site principal investigator for the finerenone trial programme (FIGARO-DKD, NEJM 2021).

Adelaide profile Google Scholar Email

Zhibin Liao

Co-lead and technical co-lead · Senior Lecturer, School of Computer Science and Mathematical Sciences · Senior Research Fellow, AIML

Computer vision researcher with more than a decade in medical AI. PhD at the University of Adelaide with Professor Gustavo Carneiro as part of the ARC Centre of Excellence for Robotic Vision, then a postdoctoral fellowship at the University of British Columbia (2017–2019) on AI for cardiovascular ultrasound. AI lead for the NHMRC RAPIDx AI partnership project (2021–2023) with Siemens Healthineers and SA Health.

Technical co-lead on the group's computer vision output, appearing in the supervising-author block on its CVPR, ICCV, ECCV and MICCAI papers, covering hallucination detection in vision-language models, counterfactual explanation, concept bottlenecks and out-of-distribution segmentation. 60+ peer-reviewed papers; 2,370+ citations, h-index 22; field-weighted citation impact 3.1 with 40% of publications in the global top 10% most cited; co-inventor on two granted US patents. Winner of the 2020 ImageCLEF VQA-Med and VQG-Med challenge. Teaches Artificial Intelligence and Computer Vision at Adelaide.

Adelaide profile Google Scholar Email

Senior collaborators and adjuncts

PhD candidates

Completed doctorates

Where alumni are now

More than twenty students and fellows have passed through the wider group since 2017, and nine doctorates have completed under the two leads. Several have gone into industry research at Google, NVIDIA and Cropify, and others to academic independence, including a continuing lectureship at La Trobe University.

Clinical and institutional collaborators

Royal Adelaide HospitalSA HealthSAHMRIJones Radiology Flinders UniversityWomen's and Children's HospitalQUT UMC GroningenSA Medical ImagingRobinson Research Institute

News

Updates, papers and appointments

All updates

Why Adelaide

People ask why a medical AI group sits in Adelaide rather than Sydney, Boston or London. The short answer is that South Australia is small enough to change and large enough to matter, and that living here is not a sacrifice you make for the work.

Aerial view of Lot Fourteen on North Terrace, Adelaide, showing the restored heritage buildings of the old Royal Adelaide Hospital site alongside new towers, with parklands behind. — Lot Fourteen on North Terrace, the innovation precinct that houses the Australian Institute for Machine Learning, where the lab is based.

8th most liveable cityin the worldGlobal Liveability Index, Economist Intelligence Unit, 2026. Adelaide was 9th in 2025.

Most affordablecapital city in AustraliaCoreLogic Quarterly Rental Review, January 2025, median weekly unit rent. Rents run about 20% below Sydney.

Most welcomingregion on EarthTop 10 Ranking, Booking.com, 2025.

The festival capitalof AustraliaAdelaide Festival, Fringe, WOMADelaide and the Cabaret Festival run through February and March.

The research argument

One health service, one state

South Australia has a single state health service covering 1.67 million people. A model that clears governance here can be deployed statewide rather than hospital by hospital, which is why three of our systems reached real clinical use.

SA Health · statewide

Everything on one street

The lab sits at Lot Fourteen, the state's innovation precinct on North Terrace, with the Royal Adelaide Hospital, SAHMRI and the university within a few minutes' walk. Clinical collaborators are a conversation away rather than a calendar invitation.

Lot Fourteen · North Terrace

Adelaide BioMed City

The precinct around us is a $3.6 billion health and life sciences cluster, among the largest in the Southern Hemisphere, bringing the Royal Adelaide Hospital, SAHMRI, the universities and the medical research institutes onto a single stretch of North Terrace.

$3.6B precinct

Australia's machine learning institute

The Australian Institute for Machine Learning is the country's largest machine learning research group, with more than 200+ NVIDIA A100 GPUs and over 5+ petabytes of secure storage available to the work.

200+ A100 · 5+ PB

Running a trial here

Prospective evaluation is where most medical AI stalls. South Australia is unusually quick to start one, which is part of why a randomised trial of clinical AI was possible here at all.

Six-week approvals

Clinical trial regulatory approval averages six weeks in South Australia, which the state puts at six to nine months saved against the alternative.

6 weeks, not 6 to 9 months

Incentives

Up to 43.5 per cent cashback is available for clinical trials, analytics, study drug manufacture and prototype manufacture.

Up to 43.5% cashback

Full trial capability

Facilities for preclinical work and for Phase One, Two and Three trials sit inside the precinct, alongside the machine learning research centres.

Preclinical through Phase 3

Precinct and trial figures from the Adelaide Economic Development Agency.

Living here

Climate and coast

A Mediterranean climate with more than 2,500 hours of sunshine a year, summers between 25 and 35 degrees, winters between 10 and 15, and no snow. More than 70 kilometres of coastline, with Glenelg, Henley, Semaphore and Port Noarlunga all reachable from the city in about half an hour. Bureau of Meteorology figures.

2,500+ hours of sun · 70 km of coast

Cost of living

The most affordable capital city in Australia. Rents are roughly 20% below Sydney, 12% below Perth and 7% below Brisbane, which on a postdoctoral salary or a PhD stipend is the difference between renting a room and renting a house. StudyAdelaide keeps a current breakdown.

~20% below Sydney

Who lives here

1.3 million people in the metropolitan area and 1.67 million across the state. A quarter of residents were born outside Australia, from more than 200 cultural and religious backgrounds, and the city has been ranked the most welcoming region on Earth. Multicultural Affairs SA and the Australian Bureau of Statistics.

25% born overseas · 200+ backgrounds

Food, wine and festivals

The Barossa Valley, McLaren Vale and the Adelaide Hills are all under an hour from the lab. February and March bring the Adelaide Festival, the Fringe, which is the second largest fringe festival in the world, and WOMADelaide.

Barossa · McLaren Vale · Adelaide Hills

Getting here and away

Adelaide Airport is fifteen minutes from the city, with direct international flights and domestic connections of 1 hour 20 minutes to Melbourne and 2 hours to Sydney.

15 min to the airport

Staying on

International graduates can access up to four years of post-study work rights, and South Australia projects 34,900 jobs of employment growth by 2029 with strong demand in health, technology and space. StudyAdelaide has the detail for international candidates.

Up to 4 years post-study work rights

Come and see

If you are considering a PhD or a postdoc with us and want to know what the place is actually like, ask. We are happy to put you in touch with people in the group who moved here from overseas.

Open positions

Join us

Join the lab

The group is co-led by a cardiologist and a computer vision researcher, so a method is assessed both on whether it beats the baseline and on whether a clinician would act on its output. Our students publish at NeurIPS, CVPR and ICCV as first authors and attend clinical meetings.

Open positions

Confirm current openings and the application address before publishing. If nothing is open, say so plainly rather than leaving a stale list.

What we look for

A strong background in machine learning or medical image analysis, evidenced by publications or a portfolio of real work.
Comfort with messy clinical data: missing labels, shifted distributions, small positive classes.
Clinicians are welcome without a coding background. Several of our projects started with a clinical question and no model.
An interest in whether the work is eventually used in practice.

How to apply

Email johan.verjans@adelaide.edu.au with:

Your CV and academic transcript.
One paragraph on why this group, specifically.
One paper of ours you have read, and what you would do differently.

Adelaide University scholarship rounds close well before intake, so start early.

Contact

Get in touch

Contact details

Address: Australian Institute for Machine Learning
Lot Fourteen, North Terrace
Adelaide SA 5000, Australia
Johan Verjans: johan.verjans@adelaide.edu.au
Zhibin Liao: zhibin.liao@adelaide.edu.au
Industry: Dr Nisha Schwarz, Business and Operations
nisha.schwarz@adelaide.edu.au
Media: Adelaide University Media Office
media@adelaide.edu.au · Newsroom

Before you email

Prospective students should read the Join us page first, which sets out what to include.

Industry partners: tell us the clinical question and what data exists, since that determines whether a project is possible. The Industry page sets out how past engagements have been structured.

Journalists: please include your deadline.