AI tool estimates melanoma depth without skin biopsy

A single number has long determined how aggressively doctors treat melanoma. Breslow depth — the measured thickness of a tumour from the skin's surface to its deepest cell — drives surgical decisions and guides whether a patient undergoes sentinel lymph node biopsy. Until now, obtaining that number required cutting into the skin.

Researchers published findings in the Journal of Investigative Dermatology describing a deep learning model that estimates Breslow depth from reflectance confocal microscopy (RCM) images alone. RCM is a non-invasive imaging technique that produces cellular-resolution pictures of living skin without removing tissue. The study reports this is the first time a computational approach has achieved both melanoma classification and numerical Breslow depth estimation through non-invasive imaging.

What the technology does

Breslow depth determines a tumour's AJCC T-category, the classification system that oncologists use to set surgical margins and decide on further intervention, according to Gershenwald and colleagues' 2017 staging guidelines. The deeper the tumour, the wider the required excision and the more likely a lymph node procedure becomes.

Previous clinical use of RCM allowed experienced operators to suspect invasion qualitatively — to recognise visual patterns associated with deeper growth — but the journal reported that a precise numerical estimate had not previously been achieved non-invasively, citing Faldetta and colleagues' 2024 review of the field.

The new model changes that, the study found, by processing RCM image stacks through a depth-aware architecture trained to map cellular features to measurable tumour thickness.

Why this matters for people with albinism

People with albinism carry significantly elevated melanoma risk. Reduced melanin means less natural UV filtration, and cumulative sun exposure translates directly into higher rates of skin cancer — particularly in sub-Saharan Africa, where dermatological care is often limited and diagnoses arrive late.

A non-invasive staging tool has specific relevance in that context. Biopsy requires trained pathologists, specialised equipment, and reliable laboratory infrastructure — resources that remain scarce across much of the region. A validated RCM-based model would not eliminate those requirements entirely, but it could allow clinicians to triage cases, prioritise interventions, and provide preliminary staging where full histopathology is delayed or unavailable.

The research is early-stage, and the journal did not report clinical trial results or regulatory approval. Independent validation across diverse skin types — including very low-melanin skin — will be a necessary next step before the tool reaches routine practice.

For now, the finding marks a meaningful shift: a number that once required a scalpel may, in time, be readable from light alone.