AI Screening and Aspirin Emerge as New Fronts in Melanoma Prevention Review

Cutaneous melanoma is the deadliest form of skin cancer, responsible for roughly 80 percent of skin cancer deaths despite accounting for only about 4 percent of skin cancer cases. A new systematic review published in Cancer Reports has pulled together the freshest evidence on how the disease might be stopped before it starts, or caught early enough to cure, and its findings point to a striking shift in the field: while traditional measures such as sunscreen and clinical examination remain the backbone of prevention, the past five years have been dominated by artificial intelligence, novel biomarkers and even an unexpected candidate for chemoprevention in the form of a humble, century-old drug.

The review, conducted according to PRISMA 2020 guidelines and registered in PROSPERO, searched Web of Science, PubMed and EbscoHost for studies published between February 2020 and January 2026. From 404 initial records, 24 studies met the strict eligibility criteria: three addressed primary prevention, meaning measures that stop the disease from developing, and 21 addressed secondary prevention, meaning screening and early detection. The studies spanned North America, Europe and Asia and included one randomized controlled trial, nine diagnostic accuracy studies, thirteen conventional observational studies and one quality improvement study. Quality was assessed with established tools including the Cochrane Risk of Bias 2.0 instrument, the Newcastle-Ottawa Scale and QUADAS-2, and the certainty of the evidence was graded using the GRADE framework.

The epidemiological backdrop gives the work its urgency. According to the Global Cancer Observatory, more than 330,000 new cases of cutaneous melanoma were reported in 2020, with nearly 59,000 deaths, translating into a mortality rate of 17.7 percent. Incidence is rising by 3 to 7 percent each year, doubling every 10 to 20 years, and the International Agency for Research on Cancer predicts an increase of more than 50 percent in new annual cases between 2020 and 2040. Fair-skinned populations in high-UV regions such as Australia, Europe and North America carry the greatest burden, and ultraviolet radiation, particularly intense intermittent exposure, remains the dominant external risk factor, driving the DNA damage and oncogenic mutations, including BRAF and NRAS alterations, that underpin melanomagenesis.

On the primary prevention front, the pickings were slim but intriguing. Two observational analyses of extended follow-up from a large Australian randomized trial, involving 19,114 participants, examined whether daily 100-milligram aspirin could prevent melanoma. In hypertensive individuals, aspirin use was associated with a 27 percent reduction in melanoma incidence, with a hazard ratio of 0.73, and the effect was strongest among those with uncontrolled hypertension. Yet in older participants overall, the same dataset showed no statistically significant benefit, with a hazard ratio of 1.04. The reviewers graded this evidence as low certainty, noting that both analyses come from the same research group and that aspirin carries well-known bleeding risks, so no clinical recommendation can be made until large, independent randomized trials clarify the risk-benefit balance.

The third primary prevention study, from Greece, took a different angle, examining total body nevus counts in 813 people. It confirmed that mole counts decline with age, with people over 51 having a 73 percent lower likelihood of carrying 30 to 60 nevi, and that taller individuals showed a steeper decline, particularly women. The finding hints at genetic mechanisms governing nevus dynamics, echoing genome-wide association work that has identified dozens of loci linking nevus count to melanoma susceptibility. Notably, the search found no new studies on sunscreen formulation or photoprotection during the review window, suggesting that after decades of successful sun-safety campaigns, research energy has migrated decisively toward early detection technology.

That migration is most visible in the eight studies evaluating artificial intelligence for melanoma detection. The review organized these tools into three generations. Classical machine learning, including support vector machines and random forests, requires experts to manually define image features. Deep learning, built on convolutional neural networks such as DenseNet and EfficientNet, learns features automatically from large labeled image sets. The newest approach, self-supervised learning, needs no labeled data at all: in a Swiss study, a self-supervised system combining object detection with vision transformers achieved 95 percent sensitivity in flagging suspicious lesions, and raised the sensitivity of medical students from roughly 63 to 68 percent up to 81 to 82 percent, matching expert dermatologists at 100 percent agreement.

Across the AI studies, specificity ranged from 70.4 to 100 percent, often matching that of specialists, but sensitivity varied widely, from 58.3 percent in two-dimensional total body photography systems to 89 percent in dermoscopic models. Three-dimensional convolutional neural networks outperformed their two-dimensional counterparts, and combining dermatologist judgment with 3D analysis improved accuracy further. Yet the review flagged consistent caveats: consumer-facing smartphone apps underperformed tools embedded in clinical settings with professional oversight; performance dropped in difficult areas such as tattoos, skin folds and hairy regions; and nearly all studies enrolled predominantly fair-skinned, Fitzpatrick type I and II patients, leaving darker phototypes dangerously underrepresented. The reviewers also raised an environmental concern, noting that the energy and water consumption of training self-supervised systems at community screening scale could be unsustainable.

Beyond AI, conventional secondary prevention showed real-world value. A five-year screening initiative in a large US health system, covering nearly 600,000 patients, diagnosed significantly more thin melanomas, those measuring 1 millimeter or less, among screened patients, while thick lesions trended downward. Free beachside screening events in Rhode Island achieved a number needed to screen of just 18.3. Longitudinal surveillance of melanoma-prone families reduced mean tumor thickness at diagnosis from 1.1 to 0.6 millimeters, with 83 percent of prospectively detected tumors classified as early-stage T1 versus 40 percent before the program. Remote training in skin self-examination proved as effective as in-person instruction in a randomized trial of 341 melanoma survivors and their partners, and community pharmacy-based teledermatology in Spain achieved more than 85 percent diagnostic accuracy on submitted photographs, opening a low-threshold referral route for the public.

Two emerging technologies rounded out the review. Hyperspectral imaging paired with deep learning distinguished melanoma from benign nevi with 98 percent accuracy using two-dimensional spectral-spatial data, exploiting spectral differences concentrated in the 500 to 675 nanometer band. Even more futuristic is a wearable microneedle patch developed in China, studded with hundreds of tiny tips and gold-silver nanospheres, that measures tyrosinase, the rate-limiting enzyme of melanin synthesis, directly within suspicious skin lesions using surface-enhanced Raman spectroscopy, with an ultralow detection limit of 0.01 units per milliliter. Because tyrosinase activity is elevated in malignant melanocytes, the reviewers suggest this painless biosensor could one day offer an objective, early biomarker, provided it remains affordable and accurate over time.

The overall message is one of cautious optimism. Advanced imaging combining three-dimensional total body photography, digital dermoscopy and reflectance confocal microscopy currently carries the highest certainty of evidence for early diagnosis, but demands specialist infrastructure. AI tools are sensitive but not yet specific enough, and must be validated across all skin tones before widespread deployment. Aspirin remains a hypothesis, not a prescription. What the review makes clear is that the future of melanoma prevention will likely be hybrid: sun protection and expert-led examination as the foundation, augmented by algorithms, biosensors and genomics, always under the supervision of dermatologists, to catch the world’s deadliest skin cancer while it is still thin, treatable and curable.

Subject of Research: Primary and secondary prevention strategies for cutaneous melanoma, including AI-based screening, chemoprevention and emerging biomarkers

Article Title: New Primary and Secondary Prevention Measures for Cutaneous Melanoma: A Systematic Review

Article References: Palau–del–Valle, M., Bendala–Tufanisco, E., & López–Ruiz, M. A. (2026). New Primary and Secondary Prevention Measures for Cutaneous Melanoma: A Systematic Review. Cancer Reports, 9(10), Article e70698. https://doi.org/10.1002/cnr2.70698

Image Credits: AI Generated

DOI: 10.1002/cnr2.70698

Keywords: cutaneous melanoma, systematic review, artificial intelligence, screening, aspirin chemoprevention, tyrosinase biosensor, total body photography, dermoscopy, UV radiation, early detection, convolutional neural networks, public health