PracticeUpdate Dermatology May 2019

EDITOR’S PICKS 11

Deep Learning-Based, Computer-Aided Classifier vs Board-Certified Dermatologists in Skin Tumor Diagnosis Take-home message • A deep convolutional neural network (DCNN) was trained using 4867 clinical images of benign and malignant dermato- logic neoplasms to determine whether this technology could be used to develop an efficient skin cancer classification system. The DCNN was then tested against board-certified dermatologists and dermatology trainees for accuracy of first-level (benign vs malignant), second-level (epithelial vs melanocytic), and third-level (specific diagnosis, such as melanocytic nevus, basal cell carcinoma) diagnosis. The accuracy of the DCNN in first-level classification was 93.4%, which was significantly higher than the board-certified der- matologists (85.3%) and dermatology trainees (74.4%). • Although, the DCNN outperformed board-certified derma- tologists in third-level classification of neoplasms as well (74.5% vs 59.7%), the authors speculate that dermatologists would likely achieve higher diagnostic accuracy if provided with more clinical information. However, these findings suggest that DCNN may play a role in aiding primary care physicians when evaluating dermatologic lesions. Caitlyn T. Reed MD Abstract BACKGROUND Application of deep-learning technology to skin cancer clas- sification can potentially improve the sensitivity and specificity of skin cancer screening, but the number of training images required for such a system is thought to be extremely large. OBJECTIVES To determine whether deep-learning technology could be used to develop an efficient skin cancer classification system with a rel- atively small dataset of clinical images. METHODS A deep convolutional neural network (DCNN) was trained using a dataset of 4867 clinical images obtained from 1842 patients diagnosed with skin tumours at the University of Tsukuba Hospital from 2003 to 2016. The images consisted of 14 diagnoses, including both malignant and benign conditions. Its performance was tested against 13 board-cer- tified dermatologists and nine dermatology trainees. RESULTS The overall classification accuracy of the trained DCNN was 76·5%. The DCNN achieved 96·3% sensitivity (correctly classified malig- nant as malignant) and 89·5% specificity (correctly classified benign as benign). Although the accuracy of malignant or benign classification by the board-certified dermatologists was statistically higher than that of the dermatology trainees (85·3% ± 3·7% and 74·4% ± 6·8%, P < 0·01), the DCNN achieved even greater accuracy, as high as 92·4% ± 2·1% (P < 0·001). CONCLUSIONS We have developed an efficient skin tumour classifier using a DCNN trained on a relatively small dataset. The DCNN classified images of skin tumours more accurately than board-certified dermatologists. Col- lectively, the current systemmay have capabilities for screening purposes in general medical practice, particularly because it requires only a single clinical image for classification. Deep-Learning-Based, Computer-Aided Classifier Developed With a Small Dataset of Clinical Images Surpasses Board-Certified Dermatolo- gists in Skin Tumour Diagnosis. Br J Dermatol 2019 Feb 01;180(2)373-381, Y Fujisawa, Y Otomo, Y Ogata et al. www.practiceupdate.com/c/79798 The British Journal of Dermatology

Risk Factors for Keratinocyte Carcinoma Skin Cancer in Nonwhite Individuals Journal of the American Academy of Dermatology Take-home message • In this retrospective chart review, squamous cell carcinoma (SCC) was the most common skin cancer in blacks and Asians and basal cell carcinoma was the most common skin cancer in Hispanics among a population of individuals who received a biopsy-proven diagnosis of skin cancer between June 2008 and June 2015. The majority of the SCC in blacks occurred in sun-protected areas (in particular, the anogenital region). Current smokers were diagnosed with skin cancer an average of 12.27 years earlier than former smokers and 9.36 years earlier than nonsmokers. • Although rates of skin cancer overall are higher in the white population, nonwhite individuals may experience greater associated morbidity and mortality. Photoprotec- tion and skin cancer screening are recommended for nonwhite patients, as is active examination of sun-pro- tected areas. Smoking cessation should be part of dermatologic counseling of all patients. InYoung Kim MD, PhD Abstract BACKGROUND As the majority of the U.S. population will consist of non- white individuals by the year 2043, it is essential that both physicians and patients are educated about skin cancer in nonwhite individuals. OBJECTIVE To update the epidemiology, investigate specific risk fac- tors, and facilitate earlier diagnosis and intervention of KC in nonwhite individuals METHODS RB-approved retrospective chart review of all non-white indi- viduals who had received a biopsy-proven diagnosis of skin cancer at Drexel Dermatology from June 2008 to June 2015. RESULTS Squamous cell carcinoma (SCC) was the most commonly diagnosed skin cancer in Black and Asian populations, while basal cell carcinoma (BCC) was the most common skin cancer in Hispan- ics. Blacks exhibited the majority of their SCC lesions in sun-protected areas, particularly the anogenital area. On average, current smokers were diagnosed with skin cancer 12.27 years earlier than former smok- ers and 9.36 years earlier than nonsmokers. LIMITATIONS Single-center design and inter-practitioner variability of skin examination CONCLUSIONS The importance of photoprotection in nonwhite individ- uals should not go overlooked. However, emphasis should also be placed on active examination of sun-protected areas in nonwhites and recognition of the relationship between HPV and genital SCC lesions. Smoking cessation should be integrated in dermatologic counseling of all patients. Interventions tailored to each of these eth- nic groups are needed. Risk Factors for Keratinocyte Carcinoma Skin Cancer in Nonwhite Individuals: A Retrospective Analysis. J Am Acad Dermatol 2019 Jan 28;[EPub Ahead of Print], KS Nadhan, CL

Chung, EM Buchanan, et al. www.practiceupdate.com/c/79417

VOL. 3 • NO. 2 • 2019