Scope of the review This paper summarizes a session from the 2017 Montagna Symposium on the Biology of Skin, which explored how large electronic registries, imaging technologies, mobile applications, crowdsourcing, and artificial intelligence are converging to transform melanoma early detection.
The case for early detection Catching melanoma early reduces the extent of surgical removal, avoids systemic therapy side effects, and dramatically improves survival. Early-stage treatment such as simple excision is also significantly less expensive than late-stage systemic regimens.
Skin's inherent advantage Unlike most cancers, melanoma arises on visible skin and mucosa, creating unique opportunities for population-wide screening, mobile imaging, and AI-assisted diagnostics that simply are not possible for internal tumors.
Schleswig-Holstein pilot A non-randomized German screening program enrolled over 360,000 residents 21 and older for total body skin examinations. During the 10-year study period, melanoma incidence rose but mortality dropped an estimated 45% for men and women in that region - a result not observed in nearby areas.
National extension falls short When the program expanded nationally, analyses produced mixed results, raising concerns about data integrity, selection bias, and birth-cohort effects. Key differences included screening age (21 versus 35 and older), discontinuation of the public awareness campaign, and altered referral pathways.
Ongoing debate The U.S. Preventive Services Task Force concluded that insufficient evidence exists to make a clear statement about skin cancer screening benefits for melanoma mortality, though it does recommend screening for higher-risk populations - highlighting the need for rigorous program design.
Population-specific design The War on Melanoma (WoM) initiative at Oregon Health and Science University adapted the German model with objective baseline metrics, control states for comparison, and risk-stratified outreach. Oregon ranks sixth nationally in melanoma incidence, making it an apt location.
Community registry Beginning in June 2014, the WoM recruited nearly 8,000 Oregonian volunteers into an IRB-approved Melanoma Community Registry. Participants commit to education, activism, and participation in research surveys and crowdsourced studies.
Grass-roots education A 'teach the teachers' curriculum provides standardized materials with competency testing for lay volunteers and skin-service professionals including massage therapists, hairdressers, and cosmetologists. A formal partnership with the Oregon Health Authority coordinates statewide release of educational materials.
App purpose and features MoleMapper is a free Apple ResearchKit app released in 2015 that lets users photographically track skin lesions over time. By including a reference coin of known size in photographs, the app calculates absolute lesion size and tracks changes quantitatively across monthly image sequences.
Clinical utility Having a longitudinal visual history available at a clinical visit enables more data-driven risk assessment. Instead of relying on a single snapshot, providers can see how a lesion has changed - a capability that could trigger earlier referral.
Open research data Users can consent to share their images for research, making curated, longitudinal datasets publicly available for AI development challenges. This addresses a critical gap: most large dermoscopy datasets remain closed to a small number of institutions.
Dermoscopy as standard of care Dermoscopy, using a magnifier (typically 10x to 20x) with polarized and non-polarized light, has become the standard initial screening tool in the U.S. It significantly increases sensitivity and specificity compared to naked-eye examination, with a number-needed-to-excise (NNE) ratio ranging from 8.7 to 29.4 depending on expertise.
Sequential digital dermoscopy For high-risk patients, digital dermoscopy documents all pigmented lesions photographically and enables objective follow-up over time. Architectural changes between visits can be detected and used to precisely triage which lesions require biopsy.
Reflectance confocal microscopy Reflectance confocal microscopy (RCM) provides painless, in vivo, cellular-level resolution imaging of the skin in approximately 5 minutes. It reduces unnecessary excisions - with NNE as low as 6.25 - and has recently gained CPT billing codes in the United States, improving its path into clinical practice.
From hand-crafted to learned programs Traditional computer vision requires experts to manually design feature detectors. Deep learning inverts this by learning relevant image features automatically from large labeled datasets, making systems more generalizable and robust.
CNN architecture and training Convolutional neural networks (CNNs) are first pre-trained on broad image datasets like ImageNet, embedding knowledge of basic visual statistics. They are then fine-tuned on domain-specific data - in this case, roughly 130,000 dermoscopic and clinical skin lesion images.
Matching board-certified dermatologists A landmark 2017 study by Esteva and colleagues showed that a CNN fine-tuned on skin lesion images matched the diagnostic performance of approximately 24 board-certified dermatologists across tasks including distinguishing melanoma from nevi and basal cell carcinoma from seborrheic keratosis. Such algorithms could be deployed on mobile devices to assist practitioners at the point of care.
Data quality and standardization Current AI systems rely on high-quality clinical images, but it remains unclear whether photos taken by lay people on mobile phones will meet the quality threshold needed for reliable diagnosis. Open, well-curated datasets are needed to address this gap.
Contextual enrichment Existing AI models classify single images in isolation. Incorporating longitudinal images, comparison with peer lesions from the same patient, and images of adjacent normal skin could shift AI from detecting whether a lesion is melanoma toward predicting whether it is likely to become melanoma.
Integration and validation The authors stress that all novel tools require rigorous validation before integration into existing health and research systems, which may be ill-prepared for the pace of technological change. Innovations must be proactively leveraged rather than reactively adopted.