Welcome back to the journal review. This is the May 2026 issue of the Journal of the American Academy of Dermatology, and we've got four pieces to get through today — a research highlights column on artificial intelligence, a commentary on a fluorescence imaging validation trial for pigmented lesions, a from-the-academy stability and safety review on in-office compounded buffered lidocaine, and a clinician's perspective piece on malpractice risk. Let's get into it.

First up is a highlights column from JAAD Reviews, written by Shari Lipner, surveying where artificial intelligence stands in dermatology right now. This isn't an original study itself — it's a curated synthesis pointing you to two underlying papers, one in JAAD proper and one in JAAD Reviews, plus some supporting literature, so I'll walk you through what she's synthesizing rather than a single methods-and-results arc.

The first underlying paper, by Kremer and colleagues, is an eye-tracking study comparing where dermatologists actually look when reading a dermoscopic image against the heat maps generated by an explainable AI algorithm called Dermoscopy EXplainable Intelligence. The clinical question here is really about trust — if an AI gives you a diagnosis, does its internal reasoning, visualized as a heat map, actually correspond to the morphologic features you as a clinician would use to make that call, or is it keying off something clinically irrelevant. What they found was a substantial pixel-wise correlation between the AI-generated heat maps and dermatologists' actual gaze patterns, and — this is the notable part — that correlation approached the level of agreement you see between two dermatologists looking at the same image independently. In other words, the algorithm's attention was landing in roughly the same diagnostic neighborhood as an expert's attention, which is a meaningful step toward interpretability rather than a black box. Lipner also folds in a related study of dermatologists using an explainable AI system to read melanoma versus nevus dermoscopic images, which showed improved diagnostic accuracy compared to a standard, non-explainable AI — reinforcing that transparency isn't just reassuring, it may actually improve performance.

The second underlying paper, by Mai and colleagues, is a systematic review of text-to-image translation algorithms — the generative models that have moved from generative adversarial networks through diffusion models to zero-shot approaches like DALL-E 2 and contrastive language-image pretraining. The application here is less about diagnosis and more about generating synthetic educational images, augmenting training datasets, and creating representations of rare disease presentations you might not have enough real clinical images of. The caveat the authors raise, and Lipner underscores it, is that these generators can perpetuate bias and lack domain-specific accuracy, so they're not ready for direct clinical integration without further validation.

The throughline Lipner draws across both papers, and a third piece of contextual literature — a systematic review of thirty-seven studies on explainable AI for skin cancer detection that found most work applies interpretability methods without actually testing whether they change dermatologist confidence or performance — is that AI in dermatology is advancing on two separate fronts: making diagnostic algorithms transparent enough to earn clinical trust, and expanding generative capabilities for education and data augmentation. Neither is plug-and-play yet. Practically, there's nothing here that changes what you do in clinic tomorrow, but it's worth knowing that explainability research is maturing to the point where eye-tracking validation against real dermatologist attention is now a benchmark, and that generative image tools, while promising for teaching and rare-disease libraries, are explicitly not validated for clinical use.

Next is a JAAD International highlights piece by Jonathan Kantor, commenting on a prospective validation trial of in vivo skin fluorescence imaging for atypical nevi and melanoma, targeting alpha-v-beta-3 integrin. This is a commentary on someone else's original study, so I'll give you the trial's shape as Kantor presents it, along with his critique.

The study, by Grossman and colleagues, enrolled two hundred forty pigmented lesions across six sites and tested various score cutoffs on the fluorescence imaging device to optimize the sensitivity-specificity tradeoff for separating minimal-risk from high-risk lesions. About half the lesions also got a dermoscopy comparison, which is what makes this interesting rather than just another device validation — you get a head-to-head against a real-world comparator that's actually used in specialized pigmented lesion clinics. At a score cutoff of seven, fluorescence imaging came out clearly ahead: sensitivity and specificity of about 89% and 94%, respectively, compared to dermoscopy's 53% and 61%. That's a meaningful gap, not a marginal one — if it holds up, it suggests fluorescence imaging could meaningfully outperform dermoscopy for triaging suspicious pigmented lesions in exactly the population where dermoscopy alone tends to struggle.

But Kantor doesn't let the promising numbers stand unchallenged. He points out that at that same cutoff, about one in ten melanomas in situ and nearly one in five high-grade dysplastic lesions scored below the threshold — meaning the device would have missed or under-called them. The authors suggest these under-scoring lesions might biologically represent slower-growing disease that could reasonably be monitored, and Kantor grants there's some biologic plausibility to that, but he's appropriately skeptical that this is settled — both clinicians and patients may have very different risk tolerance for "we'll just watch it" when the lesion in question is a melanoma in situ. He also flags the genuine advantages beyond raw accuracy: an objective, quantitative output rather than a subjective read, and real utility in triaging biopsy priority in patients with numerous clinically atypical nevi, where you're otherwise making judgment calls about which of twenty atypical-looking nevi to biopsy first. His bottom line is that this is a promising next step in the diagnostic toolkit for pigmented lesions, but it's not yet ready to change how you triage lesions in your own practice — the miss rate among high-grade lesions needs to be better characterized before anyone builds a monitoring pathway around a sub-threshold score.

Now to the piece that's most directly actionable for your day-to-day practice: the From the Academy stability and safety review on in-office compounding of buffered lidocaine and epinephrine. This is not a hypothesis-driven original study in the traditional sense — it's a formal stability and safety testing program, conducted jointly by the American Academy of Dermatology, the American College of Mohs Surgery, the American Society for Dermatologic Surgery, and the American Society for Mohs Surgery, in collaboration with the FDA, CDC, and United States Pharmacopeia, aimed at generating the evidence base for a USP monograph. So think of this less as "here's a novel finding" and more as "here's the data that will now govern what beyond-use date you're legally and practically allowed to use in your own procedure room."

The background is one every Mohs surgeon already lives with: USP General Chapter 797 currently restricts immediate-use compounded sterile preparations — which is what buffered lidocaine is, since you're mixing sodium bicarbonate into commercial lidocaine-epinephrine on site — to a four-hour beyond-use window. That four-hour rule was already loosened once, from an original one-hour limit in 2019 up to four hours in 2023, but it still creates real friction for practices that prepare buffered local anesthetic in batches ahead of a clinic day. The gap this work fills is straightforward: nobody had generated rigorous, USP-grade stability and sterility data specifically on buffered lidocaine with epinephrine to justify extending that window further.

Methodologically, this looked like formal pharmaceutical stability testing, not a clinical trial — and that's the right design choice given the question being asked. They took three production lots of buffered lidocaine nine-point-six milligrams per milliliter with epinephrine nine-point-six micrograms per milliliter, prepared in three-milliliter syringes, and stored them under two conditions — refrigerated at five degrees Celsius, and at room temperature, twenty-five degrees Celsius with controlled humidity. They then tested at time zero, six hours, twelve hours, twenty-four hours, three days, and seven days for appearance, pH, and potency and purity of both lidocaine and epinephrine by validated high-performance liquid chromatography. One lot additionally underwent particulate matter, sterility, and endotoxin testing at the start and at day seven, and container-closure integrity was tested across different needle and cap configurations, because a big part of the safety question isn't just chemical stability, it's whether the syringe-and-cap system itself maintains a sterile barrier over time. Two more lots, at different lidocaine concentrations, were used purely for antimicrobial preservative effectiveness testing. This is exactly the kind of design you'd want for a monograph-supporting dataset — regulatory-grade analytical chemistry with predefined specification limits, rather than an observational or comparative clinical design, because the question being answered is "does this preparation remain within pharmacopeial specification over time," not "does it produce a clinical outcome."

The results were reassuring across most parameters and revealing on one. Appearance stayed clear and particulate-free throughout, at both temperatures, for the full seven days. pH remained within its specified range at both storage conditions through seven days, with high statistical confidence. Lidocaine potency also held within specification at both temperatures for the full seven days. Particulates, sterility, and endotoxin all passed specification at both the initial and seven-day timepoints. The one parameter that broke was epinephrine potency at room temperature: the regression analysis showed the lower confidence bound dropping below the ninety percent specification limit at around two days when stored unrefrigerated, whereas under refrigeration, epinephrine potency held through the full seven days. That asymmetry is the clinically decisive finding here, and it's the number you actually need to remember: the assigned beyond-use date is twenty-four hours at controlled room temperature, but a full seven days if the prepared syringes are refrigerated.

There isn't a limitations section in the conventional clinical-study sense, since this is analytical stability testing rather than a comparative trial, but it's worth noting the practical boundary conditions — this data applies specifically to the tested concentrations, the specific syringe and closure system used in testing, and the specific preparation method described, not necessarily to every possible buffered lidocaine formulation or every container a practice might use.

The practical takeaway here is genuinely practice-changing, and it's the most concrete, actionable item in this episode. If you are compounding buffered lidocaine with epinephrine in-house, capped in syringes, you now have monograph-supporting evidence that room-temperature storage extends your usable window to twenty-four hours — six times longer than the old four-hour immediate-use rule — and that refrigerated storage extends it all the way to seven days. That means practices can reasonably batch-prepare buffered local anesthetic at the start of a clinic week if refrigerated, or safely prepare a day's supply the morning of, rather than mixing fresh syringes every four hours. Given how central buffered lidocaine is to Mohs and dermatologic surgery workflow, this has immediate implications for efficiency and lidocaine conservation, and it's likely to inform the actual USP monograph language going forward.

Last is a clinician's perspective piece by Warren Heymann on malpractice risk in dermatology — an editorial-style commentary rather than original research, framing three other articles in this same issue that deal with medicolegal liability, so I'll cover it the way it's written: as a synthesis and reflection rather than a study with its own methods.

Heymann opens with the framing statistic that about seven in one hundred US physicians face a malpractice claim in any given year, and lays out the three usual drivers across medicine broadly: diagnostic delay or missed lesions — especially melanoma and aggressive keratinocyte carcinomas — procedural and cosmetic complications from excisions, lasers, and fillers, and communication breakdowns, particularly failure to follow up on abnormal labs or pathology. He makes an important specialty-specific distinction here: while diagnostic failure is the leading cause of malpractice litigation in medicine generally, in dermatology specifically, injury during elective cosmetic and energy-based procedures is actually the more common trigger for a lawsuit. He cites an analysis of seventy-five liability claims tied to energy-based device procedures where laser hair removal complications — burns, scarring, dyspigmentation — accounted for almost half of all cases. Interestingly, a separate review of fifty-two lawsuits found that regardless of whether the claim was diagnostic or procedural, outcomes tended to favor the defendant dermatologist more often than not — but when plaintiffs did win diagnostic cases, the payouts were larger than in procedural cases.

He then previews three companion articles elsewhere in this same May issue. One is a retrospective review of fifty-two closed Canadian medicolegal cases, where melanoma was the diagnosis at issue in over half the cases, basal cell carcinoma and squamous cell carcinoma trailing behind, metastatic disease was present in roughly a third of cases — and melanoma accounted for the overwhelming majority of those metastatic cases — and patient death was recorded in about fifteen percent of the cohort. Another is an analysis of one hundred consecutive melanoma litigation cases, where the defense prevailed in sixty, and notably, when plaintiffs did win, pathologists were the specialty most frequently found liable, in about a quarter of plaintiff victories — a useful reminder that melanoma malpractice risk isn't contained to the clinician, it extends squarely into the dermatopathology reading. The third is a review of twenty cryotherapy-related medicolegal cases, where just over half alleged an inappropriate indication for cryotherapy — treating melanocytic nevi or unbiopsied basal cell carcinoma, squamous cell carcinoma, or inflammatory lesions without a tissue diagnosis first — and forty percent alleged injury from the procedure itself, ranging from burns and scarring to, in one case, partial nasal amputation.

Since this is a perspective piece, there's no methods or limitations section to walk through — it's Heymann's synthesis and his own risk-mitigation recommendations, which he states plainly: review and update clinical protocols regularly, keep meticulous documentation, prioritize thorough informed consent conversations, and — his personal emphasis — resist the productivity pressures of corporatized practice enough to actually sit with the patient, talk with them, and build a genuine relationship, since empathy has been shown to mediate the link between physician communication and patient satisfaction, and satisfied patients sue less. There's no new data to act on here, but the practical message worth carrying into your own practice is the specific pattern across all three companion articles: biopsy before you destroy — meaning don't cryotherapy or otherwise treat a clinically atypical pigmented or keratinocytic lesion without tissue confirmation — and recognize that melanoma-related litigation, both the diagnostic delay variety and the pathology misread variety, remains the single highest-stakes category in dermatologic malpractice, in terms of both frequency and payout severity.

That wraps up this May 2026 review — an AI landscape check that's informative but not yet actionable, a fluorescence imaging trial that's promising but needs more work on its blind spots, a stability study on buffered lidocaine that you can actually implement in your practice this week, and a malpractice synthesis worth keeping in the back of your mind every time you reach for a cryotherapy canister without a biopsy in hand. Thanks for listening, and we'll see you next month.