# JAMA Dermatology Journal Review — February 2026

Welcome back to the journal review. This is the February 2026 issue of JAMA Dermatology, and we've got two articles worth your time today — both diagnostic and technology-focused, both asking a version of the same underlying question: when we hand patients or clinicians a new imaging tool, does it actually deliver information we can trust and act on. First up is a pivotal diagnostic study on a novel cross-modal imaging device being positioned as a noninvasive alternative to biopsy interpretation. Then we'll look at a randomized study-within-a-trial comparing two tiers of patient-operated dermatoscopes for melanoma surveillance. Let's get into it.

Our first article is an original investigation, and structurally it reads like an FDA pivotal trial — because that's essentially what it is. The clinical problem is a familiar one: histopathology with light microscopy remains the reference standard for evaluating solid tissue, but it requires cutting the patient and waiting on slide processing. Noninvasive high-resolution imaging technologies — optical coherence tomography, line-field confocal OCT, high-frequency ultrasound, reflectance confocal microscopy — have been creeping into clinical use for years, but they've all shared the same limitations: grayscale, en-face orientation that takes real training to read, and bulky handheld probes that aren't especially practical chairside. The gap here is whether a newer cross-modal system — one that combines reflectance confocal microscopy with multiphoton microscopy in a cross-sectional, more histology-like display — can actually give clinicians information that lines up with real H&E and that non-expert-in-this-modality physicians can learn to read accurately. The device in question is called VIO, made by Enspectra Health, a handheld probe that scans about 0.3 millimeters into the skin using a single infrared wavelength, sorting signal into four channels that combine to approximate a familiar cross-sectional histologic view rather than the en-face view we're used to squinting at with RCM.

Methodologically, this was a prospective, multicenter, single-arm diagnostic study following the STARD reporting guideline — that's the standard framework for diagnostic accuracy studies, and choosing it signals the authors wanted this read as a rigorous accuracy trial, not just a feasibility pilot. Sixty-five adults with a lesion already scheduled for biopsy were imaged with the cross-modal device immediately before their standard-of-care biopsy, so every image had a matched histopathologic ground truth. The cohort was then split, forty percent training and sixty percent validation — a design borrowed straight from machine learning and diagnostic test validation more broadly. The logic is clean: you can't let the same readers develop their interpretive criteria and then get tested on those same images, or your accuracy numbers are inflated by memorization rather than genuine generalizable skill. So the training set was used by three expert comparative readers — dermatopathology and imaging experts — to build consensus on what primary features like epidermis, dermis, pigmented cells, collagen, and blood vessels actually look like on cross-modal images, benchmarked directly against the corresponding glass slide. Secondary features — things like stratum corneum, hair follicles, solar elastosis, hyperkeratosis, nests, atypia, epidermal disarray — emerged organically during that consensus process rather than being predefined, which the authors are transparent about.

Then came the actual test: three separate blinded physician readers, credentialed dermatopathologists or equivalent, went through about five hours of live structured training and were then tested on the validation set with zero access to clinical history, dermoscopy, or histopathology — just the cross-modal images alone. That blinded performance test is really the heart of this paper, because it's simulating the real-world scenario: can a trained physician look at this image cold and correctly call out what they're seeing.

The results were strong across the board. The three expert comparative readers reached full百分-percent consensus validating that cross-modal features did correspond to real histopathologic structures — so the ground-truth-building step succeeded completely. Then, in the blinded test, physician readers hit about ninety-six percent accuracy identifying primary tissue features, and around ninety-eight to ninety-nine percent for secondary features — both comfortably clearing the ninety-percent bar the study had set as its success threshold going in. Interrater agreement was also high, with Fleiss kappa around zero-point-nine for both region and feature calls, which in agreement-statistic terms is about as tight as you'll see in a subjective visual-read task. And on safety, there were zero device-related adverse events, with pain scores collected via the Wong-Baker scale, consistent with what you'd expect from a noncontact optical scan rather than a controlled biopsy.

For discussion and limitations, the authors are fairly restrained in their conclusions — they frame this as evidence supporting the device's FDA-cleared role in assisting clinical judgment, not replacing biopsy. Worth flagging for you as an expert audience: the cohort was overwhelmingly white, nearly ninety-nine percent, and the study excluded a fairly long list of anatomically tricky sites — palms, soles, nails, eyelids, mucosa, dense hair-bearing areas, tattoos, ulcerated or tortuous skin — meaning we have no data yet on device performance in the anatomic zones that are often the hardest ones clinically. It's also a single-lesion-per-patient design testing recognition of largely benign architectural features rather than a diagnostic accuracy study distinguishing, say, basal cell carcinoma subtype from actinic keratosis from squamous cell carcinoma in situ — this paper validates that the device shows you real histology, not that it can replace your diagnostic decision-making on ambiguous lesions.

Practically, here's how I'd frame it for you: this is an interesting, well-designed validation step, but not yet practice-changing. What it establishes is that cross-modal imaging produces images that correspond to true tissue architecture and that trained physicians — not just imaging specialists — can learn to read them accurately in a matter of hours, which lowers the barrier to clinical adoption compared to RCM's steeper learning curve. What it does not yet establish is diagnostic performance on the actual gray-zone lesions where you'd want noninvasive triage — atypical nevi, thin melanomas, BCC subtyping for treatment selection, margin assessment. If you're the kind of practice that's already exploring in vivo optical tools to reduce unnecessary biopsies or to guide biopsy site selection, this is a device worth watching as the evidence matures toward lesion-specific diagnostic accuracy data. For now, file it as foundational validation, not a change to your biopsy threshold.

Our second article is a randomized study within a trial — a SWAT, embedded inside the larger MEL-SELF randomized trial of patient-led versus clinician-led surveillance after early-stage melanoma. This is a nice example of efficient trial design: rather than running a whole separate study, the investigators piggybacked a device-comparison question onto participants who were already randomized to the patient-led surveillance arm of MEL-SELF, prespecified in the original protocol and registered in the SWAT repository. The clinical question is a practical one that's only going to become more relevant as patient-performed teledermoscopy scales: does the dermatoscope you hand your patient actually matter, or can you save money without sacrificing image quality? Specifically, they compared a lower-cost ambient-light, nonpolarized dermatoscope attachment against a pricier illuminated, polarized one — a roughly nine-fold cost difference, about thirty-five dollars versus over three hundred and twenty dollars per unit.

Two hundred fifty-one patients previously treated for stage zero to two melanoma were randomized one-to-one to the two devices, stratified by treatment center, age band, and sex. Patients used their assigned device to photograph both clinician-flagged lesions of interest and any self-detected new lesions, uploading through a teledermatology platform for remote review. This is intention-to-treat analysis, appropriately, since it's testing real-world device performance including the human factor of patients actually using these things at home. One methodological wrinkle worth noting: masking wasn't really achievable here — patients obviously knew which device they held, and the polarized device images included a visible ruler that could have tipped off the reviewing teledermatologists to allocation, even though those readers were nominally blinded. That's a real limitation the authors acknowledge, not something I'm inferring.

On the primary outcome — the proportion of participants whose baseline images were good enough for a teledermatologist to actually issue a management recommendation — the two devices performed similarly: about seventy-two percent for polarized versus sixty-eight percent for ambient-light, a difference that was not statistically significant. Extending out to twelve months, the story stayed the same, roughly eighty-one versus seventy-six percent ever receiving a recommendation, again not a significant difference. But when they drilled down to the per-image level rather than per-participant, a real signal emerged: ninety-five percent of polarized images were reportable versus ninety-one percent of ambient-light images, and that difference was statistically significant, and it was even more pronounced for self-detected lesions the patient found on their own rather than clinician-flagged target lesions. Qualitatively, teledermatologists gave more favorable feedback on polarized images and flagged blurriness and poor lighting substantially more often with the ambient-light device. Patient-reported usability, interestingly, was essentially a wash between groups — so patients didn't necessarily feel the ambient device was harder to use, even though objectively it produced more unreportable images.

The authors' own conclusion is appropriately measured: both devices enabled patients to perform dermoscopy and get a teledermatology read, and the image-quality edge of the polarized device is real but modest, and has to be weighed against that substantial cost gap. Limitations here are worth sitting with — the sample size was inherited from the host trial rather than powered specifically for this device comparison, so the participant-level primary outcome may simply have been underpowered to detect a difference that the larger per-image dataset picked up. The unmasking risk from the visible ruler is a real potential source of bias in the teledermatologist's assessment, even if it likely biases toward the polarized device rather than away from it. And this population — melanoma survivors, motivated, trained through a structured run-in with SSE videos and lesion-identification materials — may perform better with either device than an unselected, less-motivated general population would.

Practically speaking, for those of you setting up or advising on teledermatology surveillance pathways, I'd call the top-line finding — similar rates of receiving a usable recommendation — reassuring but not the full story, and the per-image reportability difference the more clinically useful signal. If you're designing a program at scale, particularly one capturing patient-initiated images of new or changing lesions rather than just a single clinician-selected target lesion, the modest but real quality gain from polarization may justify the added cost, especially since self-detected lesions — arguably the ones most likely to represent a true new primary — showed the biggest quality gap. But if resources are constrained, this study supports that the cheaper ambient-light option is a reasonable and largely equivalent starting point, not a compromise that meaningfully cripples the surveillance pathway. This is useful, actionable health-services data for anyone building or funding one of these programs, even if it's not a paper that changes how you personally read a dermoscopic image.

That wraps our two articles for this February 2026 issue. Thanks for listening, and we'll catch you next time.