How accurately can AI detect melanoma in Australia?

In Australia's SMARTI clinical trial (Victorian Melanoma Service at The Alfred hospital), the MoleMap AI algorithm achieved over 90% accuracy in identifying melanomas — a result comparable to in-person specialist assessment. A 2025 global meta-analysis of over 70,000 images found AI skin lesion diagnosis had a sensitivity of 0.91 and an AUROC of 0.88. However, the technology is still in clinical validation, and TGA-approved dermatology AI tools remain limited in number. Specificity (reducing false positives) and equity across skin tones remain open challenges.

What does a manufacturer need to do to supply an AI medical device in Australia?

Manufacturers must apply to have their product listed on the ARTG. Requirements vary by risk class (Class I to III) and must include clinical evidence, a cybersecurity plan, and a lifecycle management plan covering model updates. Existing approvals from the FDA or CE-MDR can be used as supporting evidence. The TGA's February 2026 guidance confirmed its position: AI software is assessed on its medical purpose, not the technology itself.

What is the AMA's position on AI diagnostic tools?

The Australian Medical Association (AMA) broadly supports AI as a tool to improve efficiency, but strongly emphasises that final clinical decisions must always rest with a human doctor — a principle known as clinical oversight. On the My Health Record changes, the AMA has raised concerns that patients reading technical radiology reports without medical context may misinterpret findings, leading to unnecessary anxiety or emergency visits. It has called for patients to discuss results with their GP rather than acting on them independently.

AI Diagnostic Tools in Australian Hospitals — TGA Approval, My Health Record Changes & Melanoma AI (2026)

Q: How does the TGA regulate AI medical software in Australia?

The TGA does not regulate AI as a separate category. Under its February 2026 guidance, AI software that performs a medical purpose — such as diagnosis, treatment, prediction, or monitoring — is regulated under the existing SaMD (Software as a Medical Device) framework. The deciding factor is intended medical purpose, not whether AI is involved. All such software must be listed on the ARTG (Australian Register of Therapeutic Goods) before being supplied in Australia.

Q: What is changing in My Health Record from March 2026?

From March 2026, radiology reports for X-rays of limbs (arms and legs) will be available to view in My Health Record as soon as they are uploaded. Reports for other imaging — CT, MRI, ultrasound, mammography — will have their viewing delay reduced from 7 days to 5 days. Importantly, what is shared is the written report authored by the radiologist, not the image file itself, and not an AI-generated result. From July 2026, uploading to My Health Record will become mandatory for pathology and imaging providers.

✍️ Editorial basis: This article was written by cross-referencing the TGA official website (tga.gov.au), the Australian Digital Health Agency (digitalhealth.gov.au), AMA official statements, the SMARTI clinical trial report (Journal of the American Academy of Dermatology), and a 2025 PMC open-access meta-analysis. Several claims in the original draft were found to be inaccurate or imprecise and have been corrected with sourced data. For personal medical decisions, consult your GP or contact Medicare (☎ 1800 633 422).

📋 This article is intended for general information purposes only. Health and medical device decisions should always be made in consultation with a qualified healthcare professional.

🦘 Why Australia Is Turning to AI in Healthcare

Waiting weeks to get a specialist appointment is still a familiar frustration for many Australians — particularly those living outside major cities. For some regional patients, a single dermatology visit can mean travelling hundreds of kilometres. There are only around 550 practising dermatologists across the entire country, and most are concentrated in metropolitan areas.

AI diagnostic tools have emerged as one of the most promising responses to this structural gap. In February 2026, the TGA (Therapeutic Goods Administration) updated its guidance on AI medical software regulation, and from March 2026, access to diagnostic results through My Health Record has been expanded. Together, these changes signal a meaningful shift in how Australia's healthcare system is integrating technology. This article unpacks what has actually changed — and what hasn't — based on verified sources.

📷 AI-assisted healthcare in Australia — illustrated concept (Image: Flow / AI generated)

🏛️ TGA's AI Medical Device Rules — What the February 2026 Guidance Actually Says

On 5 February 2026, the TGA simultaneously updated two complementary guidance documents on AI medical software regulation. The core message is simple but important: "AI is not regulated differently. If the software performs a medical function, the existing SaMD framework applies — full stop."

What is SaMD?

SaMD stands for Software as a Medical Device — software that performs a medical function independently of any hardware device. If an app, cloud platform, or AI system is intended to diagnose, predict, prevent, monitor, or treat a medical condition, it falls under TGA oversight. This includes AI-powered apps and chatbots. Lifestyle and general wellness apps, by contrast, are generally not regulated.

What Does Approval Actually Require?

Before being supplied in Australia, AI medical software must be listed on the ARTG (Australian Register of Therapeutic Goods). Requirements scale with risk class (Class I to III) and include: clinical evidence supporting the intended use, a cybersecurity management plan, and documentation of how AI model changes will be tracked and managed over the product's lifecycle. Existing regulatory approvals from the FDA or CE-MDR can be submitted as supporting evidence.

💡 Why This Matters
The TGA's February 2026 guidance explicitly states that "the presence of AI does not automatically increase a product's risk classification." In other words, being AI-powered doesn't trigger extra regulatory scrutiny by itself. Equally, a non-AI software tool designed for a medical purpose is just as regulated. The threshold is purpose, not technology.

📰 Source: TGA Official — AI and Medical Device Software Regulation (5 Feb 2026)

📰 Source: King & Wood Mallesons — New TGA Guidance on AI-Based SaMD (Feb 2026)

📱 My Health Record from March 2026 — What Changes and What Doesn't

This is the change that Australians — particularly patients — will feel most directly. But it's worth being precise, because the original framing of "AI sends results to your phone instantly" misrepresents what is actually happening.

What Is Actually Changing from March 2026 (Fact-Checked)

Limb X-ray reports (arms and legs): Available to view in My Health Record as soon as they are uploaded — from March 2026
CT, MRI, ultrasound, mammography and other imaging: Viewing delay reduced from 7 days to 5 days
Most blood and urine test results: Already available immediately upon upload since October 2025 — already in effect
From July 2026: Pathology and imaging providers will be legally required to upload reports to My Health Record (non-compliance will result in ineligibility for Medicare billing)

⚠️ An Important Correction
What appears in My Health Record is a written report authored by a radiologist or pathologist — not an image file, and not an AI-generated result. The doctor writes the report; the text is then uploaded. AI is not automatically interpreting and transmitting your results. The AMA has noted that because these reports are written for clinicians, patients reading them without medical context may misunderstand the findings.

📰 Source: Australian Department of Health — Modernising My Health Record: Improved Access (Jan 2026)

📰 Source: AMA — Diagnostic Imaging Reports in My Health Record: What's Changing (Feb 2026)

☀️ Australia's Melanoma Challenge — AI Skin Cancer Detection, Real Clinical Data

Australia has the highest rate of skin cancer in the world. GPs conduct over one million skin cancer consultations each year, and the chronic shortage of dermatologists — especially in regional and rural areas — makes this a natural proving ground for AI diagnostic tools.

The SMARTI Clinical Trial — Australia's Own Data

Led by the Victorian Melanoma Service at The Alfred hospital in Melbourne, the SMARTI (Skin Cancer Management with AI) trial found that the MoleMap AI algorithm achieved over 90% accuracy in identifying melanomas. This was found to be substantially equivalent to in-person specialist assessment, and comparable to specialist teledermatology review.

Associate Professor Victoria Mar, the lead researcher, stated: "With only around 550 dermatologists in Australia, most located in major cities, AI has the potential to give regional Australians equitable access to expert-level triage." That said, the technology is currently undergoing further validation in larger cohorts and has not yet been deployed broadly in GP clinics.

📰 Source: Alfred Health — Technology Fast-Tracks Early Skin Cancer Detection (SMARTI Trial)

📰 Source: Melanoma and Skin Cancer Trials — AI is a Promising Tool for Melanoma Diagnosis

Global Meta-Analysis — What the Broader Evidence Shows

A 2025 global meta-analysis published on PubMed Central (PMC), reviewing 18 studies covering over 70,000 images from 2020–2025, found that AI skin lesion diagnosis achieved a sensitivity of 0.91 and an AUROC of 0.88. However, specificity was 0.64 — meaning false positives (flagging benign lesions as suspicious) remain a significant limitation. The review also identified performance disparities on darker skin tones as an unresolved equity concern.

✅ A Note on the "85.5%" Figure
The original draft cited "85.5% accuracy at specialist level." This figure could not be traced to a verifiable source. The actual figures from Australian clinical data (SMARTI trial: 90%+) and global meta-analysis (sensitivity 0.91) are used throughout this article instead. Generalising a single paper's result as "equivalent to a specialist" risks misrepresenting a complex and still-evolving evidence base.

📰 Source: PMC Meta-Analysis — AI Skin-Lesion Diagnosis: Equity and Generalizability (Dec 2025, Open Access)

🫁 Lung Cancer AI — A Real TGA Class IIb Approval Case Study

Skin cancer is not the only area where AI is earning regulatory recognition. On 26 February 2026, the AI-powered lung nodule management solution Optellum VNC (Virtual Nodule Clinic) received TGA Class IIb approval. More than 15,000 Australians are diagnosed with lung cancer each year, yet the five-year survival rate sits at just 27%. VNC uses AI to classify and risk-stratify patients with CT-detected lung nodules, helping to prioritise high-risk cases for faster specialist review.

Optellum described the approval as "an important step toward integrating validated AI tools into routine clinical workflows — supporting, not replacing, clinical judgement." The product already holds FDA, CE-MDR and UKCA approvals, with the Australian ARTG listing representing its latest market entry.

📰 Source: PR Newswire — Optellum Secures TGA Class IIb Approval for AI Lung Nodule Solution (26 Feb 2026)

⚠️ The AMA's Warning — What to Watch Out For

The Australian Medical Association broadly welcomes AI as a tool to ease workload and improve efficiency, but has been clear about a number of practical risks that come with faster access to diagnostic information.

🔴 The AMA's Key Concerns

1. Misinterpretation of results: Radiology reports are written for clinicians, not patients. Without medical context, patients who read their own reports may misunderstand normal variants or incidental findings and make unnecessary trips to emergency departments. The AMA put it plainly: "The scan is a chapter, not the whole book."

2. Clinical oversight is non-negotiable: AI is a support tool. The final clinical decision must always rest with a qualified doctor. This is also a TGA approval condition — no AI medical software is permitted to claim it replaces clinical judgement.

3. Skin tone equity gaps: Multiple studies have identified that AI skin cancer detection tools perform less reliably on darker skin tones. This is a patient safety issue that remains unresolved in the broader rollout.

📰 Source: AMA Official — Diagnostic Imaging Reports in My Health Record: What's Changing

❓ Frequently Asked Questions

Q. How does the TGA regulate AI medical software in Australia?

The TGA applies its existing SaMD framework based on intended medical purpose — not on whether AI is involved. All medical software must be listed on the ARTG before being supplied in Australia, and must include clinical evidence, cybersecurity plans and lifecycle management documentation. This principle was reaffirmed in the February 2026 guidance update.

Q. What is changing in My Health Record from March 2026?

Limb X-ray reports will be immediately viewable, and the delay for other imaging reports drops from 7 to 5 days. What is shared is the radiologist's written report — not the image file, and not an AI-generated summary. From July 2026, uploading to My Health Record becomes mandatory for imaging and pathology providers.

Q. How accurate is AI melanoma detection in Australia?

The SMARTI clinical trial (The Alfred hospital, Melbourne) found over 90% accuracy. A 2025 global meta-analysis found a sensitivity of 0.91 across 70,000+ images. The technology is still in clinical validation and not yet broadly deployed in GP clinics. Specificity and skin tone equity remain challenges.

Q. Will AI replace my doctor?

No. TGA approval conditions, AMA policy, and clinical consensus all require that final decisions remain with a qualified human clinician. AI serves to support doctors — enabling faster triage, flagging high-risk cases, and reducing administrative burden — so that clinicians can focus on what requires their expertise.

Q. I can see my results in My Health Record — what should I do?

The AMA recommends always discussing results with your GP or specialist rather than interpreting them alone. These reports are written for clinicians and may contain unfamiliar terminology. Seeing a result appear in your record is not a reason to go to an emergency department — book an appointment with your GP to have it explained in context.

📚 References (All Freely Accessible)

All sources used in this article are publicly accessible — official government documents, clinical trial reports, and open-access research.

📄 TGA Official — AI and Medical Device Software Regulation (5 Feb 2026)

📄 TGA Official — Understanding How We Regulate Software-Based Medical Devices

📄 Australian Dept. of Health — Modernising My Health Record: Improved Access (Jan 2026)

📄 Australian Digital Health Agency — Better and Faster Access to Health Information

⚠️ AMA — Diagnostic Imaging Reports in My Health Record: What's Changing (Feb 2026)

📄 Alfred Health — SMARTI Trial: AI Skin Cancer Detection (90%+ Accuracy)

📄 Melanoma and Skin Cancer Trials — AI is a Promising Tool for Melanoma Diagnosis

📄 PMC Meta-Analysis — AI Skin-Lesion Diagnosis: Equity & Generalizability (Dec 2025, Open Access)

📄 PR Newswire — Optellum Secures TGA Class IIb Approval: AI Lung Nodule Solution (26 Feb 2026)

📄 King & Wood Mallesons — New TGA Guidance on AI-Based SaMD (Feb 2026)

📄 UQ News — PanDerm AI: 11% Improvement in Skin Cancer Diagnosis Accuracy (Jun 2025)

⚕️ Medical Information Disclaimer
This article is based on publicly available TGA guidance, AMA official statements, and published clinical research, and is intended for general informational purposes only.
It does not constitute medical advice. Individual health decisions should always be made in consultation with a qualified healthcare professional.
For Medicare or My Health Record enquiries: ☎ 1800 MEDICARE (1800 633 422)

#AustraliaHealthcare #AIdiagnostics #TGAApproval #SaMD #MelanomaAI #MyHealthRecord #DigitalHealth #AustralianHospitals #Medicare #2026MedicalTrends #PublicHealth #HealthTech