Tech Talk

Going Digital With Dermoscopy

Author and Disclosure Information

Digital dermoscopy refers to the acquisition and storage of digital images from a dermoscopic examination. In this article, we delve into the innovative world of digital dermoscopy with a review of its potential uses as well as some nuances of adapting this technology in a clinical setting, including sequential monitoring, teledermoscopy, and machine learning. We also discuss the acquisition and storage of dermoscopy images in accordance with the Health Insurance Portability and Accountability Act (HIPAA).


 

References

Dermoscopic examination has been proven to increase diagnostic accuracy and decrease unnecessary biopsies of both melanoma and nonmelanoma skin cancers.1,2 Digital dermoscopy refers to acquiring and storing digital dermoscopic photographs via digital camera, smart image capture devices such as smartphones and tablets, or any other devices used for image acquisition. The stored images may then be used in a variety of ways, including sequential digital monitoring, teledermoscopy, and machine learning.

Sequential Digital Monitoring

Sequential digital dermoscopy imaging (SDDI) is the capture and storage of dermoscopic images of suspicious lesions that are then monitored over time for changes. Studies have shown that SDDI allows for early detection of melanomas and leads to a decrease in the number of unnecessary excisions.3,4 A meta-analysis of SDDI found that the chance of detecting melanoma increased with the length of monitoring, which suggests that continued follow-up, especially in high-risk groups, is crucial.4

Teledermoscopy

Teledermatology (telederm) is on the rise in the United States, with the number of programs and consultations increasing yearly. One study showed a 48% increase in telederm programs in the last 5 years.5 Studies have shown the addition of digital dermoscopic images improved the diagnostic accuracy in telederm skin cancer screenings versus clinical images alone.6,7

Telederm currently is practiced in 2 main models: live-interactive video consultation and storage of images for future consultation (store and forward). Medicare currently only reimburses live-interactive telederm for patients in nonmetropolitan areas and store-and-forward telederm pilot programs in Alaska and Hawaii; however, Medicaid does reimburse for store and forward in a handful of states.8 Similar to dermatoscope use during clinical examination, there currently is no additional reimbursement for teledermoscopy. Of note, a willingness-to-pay survey of 214 students from a southwestern university health center showed that participants were willing to pay an average (SD) of $55.27 ($39.11) out of pocket for a teledermoscopy/telederm evaluation, citing factors such as convenience.9

Direct-to-consumer telederm offers a new way for patients to receive care.10 Some dermatoscopes (eg, DermLite HÜD [3Gen], Molescope/Molescope II [Metaoptima Technology Inc]) currently are marketed directly to consumers along with telederm services to facilitate direct-to-patient teledermoscopy.11,12

Machine Learning

Big data and machine learning has been hailed as the future of medicine and dermatology alike.13 Machine learning is a type of artificial intelligence that uses computational algorithms (eg, neural networks) that allow computer programs to automatically improve their accuracy (learn) by analyzing large data sets. In dermatology, machine learning has been most notably used to train computers to identify images of skin cancer by way of large image databases.14-17 One algorithm, a convolutional neural network (CNN), made headlines in 2017 when it was able to identify dermoscopic and clinical images of skin cancer with comparable accuracy to a group of 21 dermatologists.14 In 2018, the International Skin Imaging Collaboration (ISIC) published results of a study of the diagnostic accuracy of 25 computer algorithms compared to 8 dermatologists using a set of 100 dermoscopic images of melanoma and benign nevi.15 Using the average sensitivity of the dermatologists (82%), the top fusion algorithm in the study had a sensitivity of 76% versus 59% for the dermatologists (P=.02). These results compared the mean sensitivity of the dermatologists, as some individual dermatologists outperformed the algorithm.15 More recently, another CNN was compared to 58 international dermatologists in the classification of a set of 100 dermoscopic images (20 melanoma and 80 melanocytic nevi).16 Using the mean sensitivity of the dermatologists (86.6%), the CNN had a specificity of 92.5% versus 71.3% for dermatologists (P<.01). In the second part of the study, the dermatologists were given some clinical information and close-up photographs of the lesions, which improved their average (SD) sensitivity and specificity to 88.9% (9.6%)(P=.19) and 75.7% (11.7%)(P<.05), respectively. When compared to the CNN at this higher sensitivity, the CNN still had a higher specificity than the dermatologists (82.5% vs 75.7% [P<.01]).16 However, in real-life clinical practice dermatologists perform better, not only because they can collect more in-person clinical information but also because humans gather more information during live examination than when they are interpreting close-up clinical and/or dermoscopic images. In a sense, we currently are limited to comparing data that is incommensurable.

Machine learning studies have other notable limitations, such as data sets that do not contain a full spectrum of skin lesions or less common lesions (eg, pigmented seborrheic keratoses, amelanotic melanomas) and variation in image databases used.15,16 For machine algorithms to improve, they require access to high-quality and ideally standardized digital dermoscopic image databases. The ISIC and other organizations currently have databases specifically for this purpose, but more images are needed.18 As additional practitioners incorporate digital dermoscopy in their clinical practice, the potential for larger databases and more accurate algorithms becomes a possibility.

Image Acquisition

Many devices are available for digital dermoscopic image acquisition, including dermatoscopes that attach to smartphones and/or digital cameras and all-in-one systems (eTable). The exact system employed will depend on the practitioner's requirements for price, portability, speed, image quality, and software. Digital single-lens reflex (DSLR) cameras boast the highest image quality, while video dermoscopy traditionally yields stored images with poor resolution.19 Macroscopic images obtained by other imaging devices, including spectral imaging devices and reflectance confocal microscopy, usually are yielded via video dermoscopy or a video camera to capture images; thus, stored images generally are not as high quality.

Smartphones are increasingly used for clinical imaging in dermatology.20 Although DSLR cameras still take the highest-quality images, current smartphone image quality is comparable to digital cameras.21,22 Computational photography uses computer processing power to enhance image quality and may bring smartphone image quality closer to DSLR cameras.22,23 Smartphones with newer dual-lens cameras have been reported to further improve image quality.21 Current smartphones have the option of enabling high-dynamic-range imaging, which combines multiple images taken with different exposures to create a single image with improved dynamic range of luminosity. It has been reported that high-dynamic-range imaging may even enhance dermoscopic features of more challenging hypopigmented skin cancers.24

Next Article:

Clinical Pearl: Mohs Cantaloupe Analogy for the Dermatology Resident

Related Articles