⚡New AI Model for Dermatology

Good morning — it’s Saturday, July 26, and we’re covering a first-of-its-kind multimodal AI model in dermatology, a landmark randomized trial in lentigo maligna, and more.

Quick heads-up. We’re off next week, back in your inbox the following Saturday.

First time reading? Sign up here.

Enjoying the newsletter? Share it with a friend or colleague — here’s the link.

P.S. You can send me feedback at [email protected]

A study published in Nature Medicine introduces PanDerm, a first-of-its-kind multimodal dermatology foundation model trained on more than 2 million images across four modalities: clinical photos, dermoscopy, total-body photography, and dermatopathology.

Developed using self-supervised learning across data from 11 institutions, PanDerm addresses a key limitation of existing AI models—their inability to integrate the multiple imaging types central to dermatologic diagnosis.

In validation, PanDerm delivered state-of-the-art performance across 28 clinical tasks, from skin cancer screening and metastasis prediction to diagnosis of rare and common skin diseases. It often outperformed existing models using just 10% of labeled training data.

Three clinician reader studies demonstrated PanDerm’s real-world potential.

The first study found that PanDerm outperformed seven dermatologists and five trainees in detecting early-stage melanoma from serial dermoscopic images. At the first follow-up time point, it correctly identified 78% of melanomas, compared to 33% by the human reviewers.

The second study involved 41 clinicians who assessed dermoscopic images of pigmented lesions before and after assistance from PanDerm. Diagnostic accuracy improved from 69% to 80%.

The final study evaluated PanDerm’s ability to support differential diagnosis across 128 skin conditions using clinical photographs. After reviewing PanDerm’s output, the proportion of readers who included the correct diagnosis among their top three differentials rose from 51% to 68% among generalists and from 55% to 65% among dermatologists.

Further real-world testing is planned before clinical implementation.

> Radiotherapy and imiquimod both clear lentigo maligna, though imiquimod trends toward fewer long-term treatment failures: In the first randomized trial comparing different treatment modalities for lentigo maligna, both radiotherapy and topical imiquimod achieved 95% biopsy-confirmed clearance at 6 months. The findings come from a multicenter Phase 3 study published in JAAD. Investigators randomized 118 patients who were ineligible for surgery to either radiotherapy or 12 weeks of imiquimod. All patients underwent targeted biopsy at 6 months to assess for residual disease, with further clinical follow-up through 24 months. Although treatment failures at 24 months occurred less frequently with imiquimod (10% vs 21%), the study was underpowered to detect a statistically significant difference (OR 2.35, 95% CI 0.82–6.75; P = 0.11). No cases of invasive melanoma were observed. Both treatments were well tolerated; most skin reactions were mild to moderate. Study limitations included a lack of blinding and a smaller-than-planned sample size due to COVID-19–related recruitment challenges. Radiotherapy provides structured, supervised care but requires daily clinic visits over 5 to 6 weeks. Imiquimod allows at-home treatment, though patients often need dose adjustments to maintain inflammation without causing discomfort.

> Baricitinib plus 308 nm excimer laser delivers strong repigmentation in localized refractory vitiligo: Findings from a randomized trial highlight the additive benefit of combining oral baricitinib with the 308 nm excimer laser in localized, treatment-refractory non-segmental vitiligo. A total of 251 adults with vitiligo affecting less than 10% of body surface area and unresponsive to conventional therapy were randomized into four groups. At Week 52, 100% of patients in the baricitinib plus excimer laser group achieved at least a 50% improvement in vitiligo severity, measured using the Vitiligo Area Scoring Index (VASI). In comparison, this threshold was reached by 90% on baricitinib alone, 77% on topical tacrolimus plus excimer laser, and 15% on excimer laser alone (P < 0.001). Facial lesions consistently responded better than non-facial sites across all treatment arms. One year after treatment cessation, disease stability—defined as a ≤10% change in VASI—was highest in the combination group (97%) and lowest in the excimer laser monotherapy group (82%). No serious adverse events or treatment discontinuations were reported. Notably, a pigment island density of ≥5/cm² at Week 4, identified by dermoscopy, was associated with a 10% increase in repigmentation, suggesting its potential as an early marker of response. Study limitations include small group sizes and a lack of blinding.

> ChatGPT and Heidi outperform traditional dictation in generating dermatology consultation letters: AI-generated consultation letters may offer faster, more accurate documentation for dermatologists, according to a pilot study published in Clinical and Experimental Dermatology. Using four scripted dermatology cases enacted by two clinician–patient pairs, researchers compared consultation letters produced through traditional clinician dictation and secretary transcription to those generated by two AI platforms: ChatGPT and Heidi. Both AI tools generated letters by transcribing and summarizing recorded consultation audio. All letters were assessed based on their initial output, without post-editing or clinician-led proofreading. ChatGPT and Heidi performed the best across accuracy (93% vs 85%), user satisfaction (4.6 vs 4.3/5), and generation time (54 vs 27 seconds). Heidi, a medical-domain AI model, produced more structured output and encountered fewer technical issues than ChatGPT. Traditional letters were easier to read but scored lower in accuracy (59%), satisfaction (3.5/5), and speed (10 minutes 54 seconds). Limitations include the small sample size (four cases) and the use of scripted scenarios rather than real-world consultations. The authors recommend further testing in clinical settings to assess utility.

> Topical botanical shows efficacy in moderate-to-severe pediatric alopecia areata: The RAAINBOW Phase 2/3 trial suggests that Cinainu—a topical formulation derived from four botanical extracts—may provide a safe and effective treatment option for children with moderate-to-severe alopecia areata. Cinainu is thought to act by modulating IL-8 signaling, reducing proinflammatory adhesion molecules, and restoring expression of the perifollicular anti-apoptotic protein Bcl-2. The study, published in the British Journal of Dermatology, included 62 patients aged 2–17. They were randomized to Cinainu or placebo for 24 weeks, followed by a 24-week treatment-free follow-up. Mean scalp hair coverage at baseline was 42%. At Week 24, 26% of patients treated with Cinainu achieved at least 80% scalp hair coverage (SALT score ≤20), compared to 10% on placebo (P = 0.14). By Week 48, these figures increased to 48% and 15%, respectively (P = 0.01). Adverse events were generally mild or moderate. Study limitations include the small sample size and attrition over time: 84% of patients completed the treatment period, but only 68% completed the full study. Treatment options for alopecia areata remain limited. There are no approved therapies for moderate disease, topical use, or pediatric patients—aside from ritlecitinib, which is approved for severe disease in adolescents.

Quizzes, cases, and perspectives to sharpen your clinical eye.

The US FDA and UK MHRA have approved sebetralstat (Ekterly), the first oral on-demand treatment for hereditary angioedema, based on data from the Phase 3 KONFIDENT trial. The plasma kallikrein inhibitor showed faster time to symptom relief, reduction in attack severity, and complete resolution compared to placebo (More)

Switching to the IL-13 inhibitor tralokinumab led to a ≥75% improvement in eczema severity (EASI 75) in 75% and 83% of patients with atopic dermatitis who were unresponsive to upadacitinib 15 mg and 30 mg, respectively, according to a 48-week study of 24 patients (More)

In a Japanese study of four patients with recessive dystrophic epidermolysis bullosa, weekly topical B-VEC led to complete closure of all treated primary wounds by Months 3 and 6, with 75% of those wounds remaining closed at Month 12. B-VEC is an HSV–1–based gene therapy that delivers COL7A1 directly to wounded skin (More)

A meta-analysis of 12 studies (N=2,606) found that patients with psoriasis but no musculoskeletal symptoms were 2.5 times more likely than healthy controls to show subclinical synovitis on ultrasound or MRI (RR 2.55; 95% CI, 1.18–5.52) (More)

A retrospective study of 140 patients classified hand vitiligo into four distributional subtypes: focal/scattered (46%), distal digit (31%), universal (13%), and proximal digit (9%). The focal/scattered type predominated in pediatric cases (69%), while the universal type had a later onset and poorer treatment response (More)