DermaSensor

The model is the product. The handheld device shines broadband light on a lesion and records how it scatters beneath the surface; a spectrum carries no reading a physician could apply, and a machine learning algorithm trained on spectra at 47 selected wavelengths turns five recordings into a Monitor or Investigate Further result with a 1 to 10 similarity score. There is no mode in which the optics alone give a clinical answer, so without the model there is nothing to sell.

What the device may and may not do is published, with the output scale and what it means. The labeling says it is an adjunct for physicians who are not dermatologists, for lesions already suspicious for melanoma, basal cell or squamous cell carcinoma, in patients 40 and over; it is not a screening tool, not a sole diagnostic criterion and not a confirmation of diagnosis. It is untested on lesions under 2.5 or over 15 millimeters and excluded on palms, soles, mucosa, genitals, ears, near the eye and under nails, and physicians are told to refer pigmented lesions in darker skin types primarily on clinical concern. The result is Monitor or Investigate Further, with a 1 to 10 similarity score whose cancer probability is published, and the device recalibrates in its base before each patient.

The reader studies measured how physicians' referrals changed with the device, including where they changed for the worse. The decision summary's finding that negative results repeat poorly between scans is not reflected in the FAQ's description of a Monitor result.

The approach and its training data are described; the model and its update practice are not. The FDA decision summary sets out the development: elastic scattering spectra from 360 to 810 nanometers reduced to 47 wavelengths, an algorithm trained on 950 lesions (about 4,200 spectra) and tuned on separate study data, with the demographics and body sites of both sets tabulated. The FAQ calls the algorithm locked and non-adaptive.

The model type is not named, and nothing published says how a new version would be validated or when customers would receive it, although the rental includes software updates. The workflow page's description of an algorithm trained on more than 20,000 scans from over 4,000 lesions does not match the decision summary's training set.

A short chain at the model layer: the algorithm is DermaSensor's own, locked, and runs on the device. No outside model interprets the spectra, and the training data came from the company's own clinical studies under trial protocols, described in the FDA decision summary.

The parties around the device are not named: the cloud provider that stores results, any subprocessors, and whoever supports remote monitoring. Whether commercially captured scans are used for future submissions is not stated.

A prospective, blinded multicenter study supports the device, and it shows high sensitivity bought with low specificity. DERM-SUCCESS enrolled 1,005 patients with 1,579 lesions at 22 sites, 18 in the United States, with biopsy as the reference: sensitivity for high-risk lesions was 95.5 percent against 83.0 percent for primary care physicians, specificity 20.7 percent against 54.2 percent, and the area under the curve 0.79 against 0.74. Reader studies with more than 280 primary care physicians found aided physicians referred more of the cancers; in the melanoma reader study, decisions corrected by the device outnumbered those it made worse by 248 to 49.

The reader studies used images rather than live patients, the pivotal cohort was 97 percent White, and no study yet shows earlier diagnosis or better outcomes in practice.

Two facts about the data path are published, and nothing on retention or use. The FAQ says the algorithm on the device is locked and non-adaptive, so scans do not retrain it in use, and that the monthly rental includes cloud storage for device results and remote device monitoring by the company. What the cloud copy contains (the score alone, or patient identifiers and lesion location), how long it is kept, who can see it, and whether commercial scans feed future versions are not stated. The privacy policy (effective 14 February 2026) covers website and account information, not patient results.

The device records light scattering rather than photographs, so no skin image is created. Safety engineering goes undescribed as well: there is nothing on guardrails, hallucination mitigation or how a safety event is handled.

No status is stated, and no privacy document reaches the product. The privacy policy (effective 14 February 2026) covers website, account and marketing information and does not mention patient results or HIPAA, and the website terms date from 2020. The rental includes cloud storage of device results, so patient-linked data may reach the company, but no business associate position, agreement or statement of what the stored results contain is published.

An outside check of the device software, and nothing on the company or its cloud. The FDA decision summary records that the De Novo submission included cybersecurity documentation (a hazard analysis covering unauthorized access, mitigations and verification testing) reviewed under FDA's premarket cybersecurity guidance. No SOC 2 report, ISO/IEC 27001 certificate, trust portal or vulnerability disclosure policy is published, and nothing addresses the cloud that stores results or the remote monitoring connection.

The authorization is identifiable, with its indication and controls. FDA granted De Novo DEN230008 in January 2024, creating product code QZS for a software-aided adjunctive diagnostic for physicians on lesions suspicious for skin cancer, Class II with special controls that include a post-market study over-representing darker skin types. The indication is lesions suggestive of melanoma, basal cell or squamous cell carcinoma in patients 40 and over, to assist the decision to refer to a dermatologist, used with the full clinical assessment by physicians who are not dermatologists.

The company's site describes the device as FDA cleared; it was granted De Novo authorization, which later devices of the type can cite as a predicate. It also holds a CE mark.

Subgroup performance is published in the FDA decision summary, and it shows where the evidence is thin. Tables by Fitzpatrick skin type and age give device sensitivity and specificity with counts, and the reader study's effect on physicians in each group. Melanoma data in darker skin types is sparse: types IV to VI contributed 12 melanomas in the pivotal study (device sensitivity 10 of 12 against 8 of 12 for unaided physicians), and in the pigmented lesion reader study, aided physicians did worse than unaided for type IV (57.8 against 72.7 percent, on two lesions). Specificity falls with age, to 12.6 percent at 70 to 79 and 10.8 percent at 80 to 89. The training data was mostly older, type I to III and White patients, and the pivotal cohort was 97 percent White.

The labeling tells physicians to base referral of pigmented lesions in types IV to VI primarily on clinical concern, and FDA's special controls require a post-market study that over-represents darker skin types. The company's statement that performance is consistent across skin tones goes further than the decision summary does.

Error rates are published with their method and denominators, and nothing stands behind them. In the pivotal study the device's sensitivity for high-risk lesions was 95.5 percent against 83.0 percent for primary care physicians, and its specificity 20.7 percent against their 54.2 percent, so most benign lesions scanned return Investigate Further; the FAQ gives the probability of cancer by similarity score, from 6 percent at 1 to 61 percent at 10. The decision summary also reports low repeatability of negative results and of the similarity score between repeat scans of the same lesion, while positive results repeated more than 90 percent of the time.

The website terms (last updated August 2020) disclaim any warranty on accuracy and have the user indemnify the company; the device rental terms are not published, and no guarantee or remediation attaches to a result.

No integration exists. The FAQ answers the question directly: the device is not presently integrated with any EHR; practices upload an electronic copy of the DermaSensor report to the patient's record or type the 0 to 10 result into the note by hand. The company says integration will follow depending on customer needs.

The device runs on its own at the point of care, and the cloud side is unlocated. The algorithm runs on the handheld device and returns a result on screen within seconds, so the clinical function does not depend on a connection. The rental includes cloud storage of results and remote device monitoring by the company, but the hosting provider, the region where results are stored and what the monitoring connection carries are not stated. The device is sold in the United States and carries a CE mark.

No price is published, and how the device is sold is discoverable. The FAQ says the device is rented monthly, with packages set by anticipated use, on a one-year agreement that can be ended after the first year; the rental includes software updates, support, replacement devices, cloud storage of results and remote monitoring, and onboarding and training are included. There is no published price for the rental or for buying outright, and pricing starts with a demo request.

On billing, the FAQ says no dedicated CPT code exists, physicians have billed under unlisted codes, some payers treat the service as experimental, and patients may pay out of pocket; a coding guide and reimbursement hotline are offered. That is payer context rather than a price.

Deliberately narrow, and validated where it is labeled to work. The device is for physicians who are not dermatologists, mainly in primary care, on lesions they already find suspicious, in patients 40 and over, with lesion size and body-site exclusions stated. That is the setting of the pivotal study, which enrolled at 22 sites with primary care physicians, and of the reader studies with more than 280 primary care physicians.

The labeling is candid about the thinnest part of the evidence, melanoma in darker skin types, and FDA requires a post-market study there. The company also markets to specialty, value-based and membership-based practices; the indication excludes dermatologists.