Cutaneous neoplasias in epidermolysis bullosa: molecular characterization and application of augmented intelligence
Final Report Abstract
Epidermolysis bullosa (EB) is a group of rare skin diseases with increased skin fragility caused by genetic defects in the skin’s structural proteins. Blisters and wounds occur spontaneously or after minor trauma. As a severe complication, squamous cell carcinoma (SCC), an aggressive skin cancer, can occur from adolescence onwards. Early recognition of SCCs is difficult, but crucial for therapy, as good therapeutic options for advanced SCCs are amiss. In project part A, we developed an artificial intelligence (AI) model to recognize SCCs on photos of EB skin. Hundreds of photographs of RDEB SCCs and non-SCCs were collected through an international expert network specifically set up for this purpose. A deep learning AI model was trained on the images, and clinical data went into a second model. The resulting model is a binary classifier with a confidence interval between 0 and 1 as its output (0= favoring no SCC, 1=favoring SCC). The model’s accuracy was above the joint performance level of 21 international EB experts. Our goal is to make the model accessible to physicians and patients via an app. In project part B, we analyzed EB nevi. These are characteristic, large pigmented sports occurring in at least 14% of EB patients. They can grow rapidly and are often clinically atypical, thus exhibiting certain characteristics of melanoma. The exact pathogenesis is unclear, as is the question whether these are real nevi with a potential for malignant transformation through mutations in the BRAF and NRAD genes. We have accrued the largest cohort of EB nevi to date through an international collaboration, comprising 369 nevi in 170 patients and, where available, tissue data and material. We found that EB nevi can appear at any age, but do so in median at 11 years. They are lighter and larger than “normal” nevi and occur evenly on all body parts, while “normal” nevi favor the back. More than 70% of EB nevi for which this information was available change over time (e.g. in size, shape, color, or they regress and vanish). Only 2.8% of EB nevi were biopsied or excised and none was malignant. Tissue samples were available for 8 nevi, of which 7 showed genetic changes is BRAF or NRAS. Findings from both project parts can potentially improve counselling and clinical care for people with EB. At the same time, the use of global expert networks and advanced AI models can serve as a role model for other rare sin diseases.
Publications
-
Gastrostomy tube feeding in epidermolysis bullosa: A multi‐center assessment of caregiver satisfaction. Pediatric Dermatology, 40(2), 270-275.
Kleinman, Elana P.; Reimer‐Taschenbrecker, Antonia; Haller, Courtney N.; Paller, Amy S.; Levy, Moise L. & Eichenfield, Lawrence F.
-
A call for implementing augmented intelligence in pediatric dermatology. Pediatric Dermatology, 40(3), 584-586.
Issa, Christopher J.; Reimer‐Taschenbrecker, Antonia & Paller, Amy S.
-
Oral presentation in late-breaking research session of the American Academy of Dermatology (AAD) on 3/18/2023: Squamous cell carcinoma image recognition in recessive dystrophic epidermolysis bullosa: Developing a first-in-genodermatosis deep learning model
Reimer-Taschenbrecker A., Adate A., Furmanchuk A., Katsaggelos A.K., Temps W.H., Meredith M.A., Hwang A., Bhuva D., Nardone B., Kho A. & Paller A.S.
-
Osteoporosis and bone health in pediatric patients with epidermolysis bullosa: A scoping review. Pediatric Dermatology, 41(3), 385-402.
Kwon, Andie; Hwang, Austin; Miller, Corinne H.; Reimer‐Taschenbrecker, Antonia & Paller, Amy S.
-
Therapies for cutaneous squamous cell carcinoma in recessive dystrophic epidermolysis bullosa: a systematic review of 157 cases. Orphanet Journal of Rare Diseases, 19(1).
Hwang, Austin; Kwon, Andie; Miller, Corinne H.; Reimer-Taschenbrecker, Antonia & Paller, Amy S.
