Early diagnosis is critical in all forms of cancer. Skin cancer is the most common form of cancer diagnosed in the U.S., and when it’s caught early, it is very treatable. However, sometimes getting an appointment with a dermatologist can take months. Several studies have validated that many patients have long wait-times before they can get in to see a dermatologist, even those with urgent problems, such as a mole that has changed in color or size.1
The use of technology may make diagnosis of skin cancers easier. Some researchers are studying the use of technology, including artificial intelligence to identify skin cancers like melanoma. One of the most promising uses of technology in skin cancer detection has been developed by a group of researchers at Stanford University. The team used deep convolutional neural networks (CNNs) to classify skin lesions. CNNs are the artificial intelligence behind many image algorithms online, like the technology Facebook uses to identify who to tag in your photos and Google uses for image searches. CNNs are trainable – by feeding them images and telling them what the image means, the technology learns over time to recognize similar images.2,3
The Stanford research team, led by Andre Esteva, conceived the idea of using CNNs because most skin cancers are diagnosed visually, either found by the individual during a self-examination or during a clinical exam by a doctor. While classification of skin lesions using images can be a challenging task, CNNs have potential for highly variable tasks.
Esteva and his team trained a CNN using 129,450 clinical images of 2,032 different dermatological diseases. They then tested the CNN against 21 board-certified dermatologists on images of lesions that had been proven through follow-up biopsies. The CNN and dermatologists were tested on two classification categories: 1) keratinocyte carcinomas versus benign seborrheic keratoses, and 2) malignant melanomas versus benign nevi (moles). The first category represented the most common skin cancers (keratinocyte carcinomas), and the second category represented the deadliest form of skin cancer (melanoma).
The CNN was as good as the expert dermatologists in identifying the lesions correctly in both categories. The results were published in the journal Nature. This was an exciting development for using technology to expand the capability of dermatology specialists and potentially triage patients, identifying those who need further follow-up versus those who have benign (non-cancerous) lesions. However, additional research is needed to evaluate its performance in real-world settings, as well as validating the technique across a spectrum of skin diseases that are seen in clinical practice.2
Is there an app for that?
Naturally, this kind of technology could be used in smartphone applications (“apps”), and with a projected 6.3 billion smartphone subscriptions estimated to exist by the year 2021, use of technology to identify skin cancers could potentially provide low-cost universal access to vital diagnostic care.2
Some smartphone apps already exist that focus on skin cancer. An analysis of several smartphone apps found that most focus on education about melanoma, ultraviolet (UV) radiation exposure prevention, and skin self-examination strategies using the ABCDE method (A=Asymmetry; B=Border; C=Color; D=Diameter; E=Evolving). Some apps also provide the ability to store images of skin lesions for later review by a dermatologist or for self-monitoring to evaluate changes over time. A few apps offer expert review of images, usually at an additional fee. However, none of the apps evaluated have been validated for diagnostic accuracy using established research methods.4
The validation is important because of the risk of false negatives. False negatives occur when a test or application says you do not have a condition when you actually do. This could be concerning in the case of a possible skin cancer, when early detection is so critical. While the advances in technology are exciting and promising, for now, your best bet is still monthly self-examinations and regular clinical exams by a doctor.
Tsang MW, Resneck JS Jr. Even patients with changing moles face long dermatology appointment wait-times: a study of simulated patient calls to dermatologists. J Am Acad Dermatol. 2006 Jul;55(1):54-8.
Esteva A, et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017 Feb;542:115-118. doi:10.1038/nature21056.
A Beginner’s Guide to Understanding Convolutional Neural Networks. Accessed online on 5/15/17 at https://adeshpande3.github.io/adeshpande3.github.io/A-Beginner%27s-Guide-To-Understanding-Convolutional-Neural-Networks/.
Kassianos AP, Emergy JD, Murchie P, Walter FM. Smartphone applications for melanoma detection by community, patient and generalist clinician users: a review. Br J Dermatol. 2015 Jun;172(6):1507-18. doi:10.1111/bjd.13665.