Dermatologists should lead efforts to elucidate these and other issues pertaining to the optimal use of deep learning in skin cancer screening. One of the first tasks toward this goal should be “defining thresholds for probability of malignancy that should prompt the patient to immediately see a dermatologist.”3 The CNN models generate predictions as a probability distribution of diagnoses, which a patient or non-expert clinician would have difficulty interpreting.

Based on the findings of these emerging studies, the potential value of deep learning models in skin cancer detection is clear. “Without the leadership of dermatologists, however, the tremendous potential of deep learning to change the field may never be fully achieved,” Zakhem et al, concluded.3

Dermatology Advisor checked in with one of Zakhem’s co-authors, Roger S. Ho, MD, MS, MPH, FAAD, assistant professor of dermatology in the Ronald O. Perelman Department of Dermatology at the NYU School of Medicine, to further discuss the use of AI for skin cancer detection .

Dermatology Advisor: What are some of the remaining barriers in the development of AI for skin cancer detection?

Dr Ho: There are still many foreseeable barriers in achieving 100% AI accuracy in skin cancer detection. Both physicians and patients need to understand that AI currently cannot detect skin cancer with 100% accuracy, and this may not be achievable in the foreseeable future because of myriad logistical, methodological, and technological barriers. The algorithm is only as good as we train it to be, and we currently do not have the repertoire of images needed that are representative of all skin cancers, all skin types, all ethnic groups, all age groups, all skin anatomic locations, and under all conceivable permutations of image capturing conditions, lighting, and devices to train it perfectly.

Dermatology Advisor: Considering these challenges in achieving complete accuracy, what is the potential role of AI in skin cancer diagnosis?

Dr Ho: AI technology does have a role in helping clinicians when they are unsure about the behavior of a specific lesion; it can help them expand on their differential diagnosis, help with triaging, and guide them on the need for referral or further diagnostic testing such as a skin biopsy. Ultimately, this technology will help prioritize resources for patients and streamline patient access and patient education, and the dermatological community is committed to working with different technological platforms to enhance the algorithms in achieving this.

Dermatology Advisor: What additional points would you like to share with clinicians regarding this topic?

Dr Ho: AI technology is not meant to be used to replace clinicians. However, when used in conjunction with guidance from clinicians, AI technology has the potential to transform skin cancer screening and provide more efficient and precise dermatologic care.

Follow @DermAdvisor


1. Han SS, Kim MS, Lim W, Park GH, Park I, Chang SE. Classification of the clinical images for benign and malignant cutaneous tumors using a deep learning algorithm. J Invest Dermatol. 2018;138(7):1529-1538.

2. Esteva A, Kuprel B, Novoa RA, et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017;542(7639):115-118.

3. Zakhem GA, Motosko CC, Ho RS. How should artificial intelligence screen for skin cancer and deliver diagnostic predictions to patients? JAMA Dermatol. 2018;154(12):1383-1384.