The role of Artificial Intelligence in dermatology
An overview of how Artificial Intelligence can improve diagnoses and offer more timely and effective care to patients.
Artificial Intelligence in Dermatology
Artificial intelligence (AI) is a term that refers to the development of computer systems to perform tasks normally attributed to human intelligence. To do this, technologies such as neural networks are used to imitate the functioning of the human brain. In narrow applications such as playing chess or Go, AI now outperforms humans. Recently, dramatic advances in machine learning, which is a subset of AI, have made it possible to apply AI to certain aspects of everyday dermatology.
Applications of Artificial Intelligence in dermatology
Most applications of AI in dermatology focus on using machine learning (ML) to differentiate between benign and malignant skin lesions. Since ML is a generic technology, it can be used for non-malignant disorders as well. Reported applications include assessment of skin ulcers and prediction of psoriasis’ response to biologics. In dermatological research, ML has been applied to areas such as using gene expression profiles in disease classification (Gomolin et al. 2020).
Despite the evident promise of AI in dermatology, barriers to its wider adoption remain. These include difficulties in standardization, acceptance by doctors and patients, the availability of training data and liability (Gomolin et al. 2020). Clearly, there is a need for more research in this field to identify the best use cases, confirm the validity of the technologies and provide the scientific basis for their wider adoption. A recent survey “Attitudes towards artificial intelligence within dermatology: an international online survey” of 1,271 dermatologists found that 85.1% of respondents were aware of AI as an emerging topic in their field, but only 23.8% considered themselves as having good or excellent knowledge of the subject (3). Moreover, 77.3% agreed that AI will improve dermatological treatment and 79.8% thought that AI should be part of medical training.
The COVID-19 pandemic has accelerated the adoption of AI because it can help to reduce the need for direct patient/doctor contact. However, there is still not enough data to support the effectiveness of using AI in dermatological diagnosis. It is true that there are a number of apps that claim to be able to diagnose, monitor and treat a range of skin disorders, but a systematic review has revealed serious diagnostic failures.
During the Google I/O Developer Conference in May 2021, a “dermatology assistant” tool was presented that uses the phone camera to help detect and classify skin, scalp and nail problems. The tool has not yet been evaluated by the FDA. However, it has been CE marked as a Class I medical device in the European Union. This means that it is not considered as a substitute for medical advice, but can support clinical decision making.
The University of Gothenburg in collaboration with the Sahlgrenska University Hospital, published an article in the Journal of the American Academy of Dermatology entitled “Discrimination between Invasive and In situ Melanomas Using a Convolutional Neural Network“.
The article reports on a study that attempted resolve a challenge in the dermatoscopic diagnosis of suspected melanoma: deciding whether it is in situ or invasive. The researchers collected a set of 937 dermatoscopic photographs made between 2016 and 2020, that were used to train a convolutional neural network (CNN). This is a special class of artificial neural network (ANN), a computing system that is loosely based on the structure of animal brains. As do the latter, an ANN has input, computation and output layers that process information and present the result. CNN are particularly suited to image processing and for that reason are often used in dermatology.
In this instance, the CNN came close to the performance of a panel of seven dermatologists in terms of accuracy and speed. When compared to individual panel members, the CNN has similar performance. Given that results were obtained with only a small number of images, we can expect continued development to improve on the outcomes. The end goal is to develop a system that can support patient and doctor by providing an accurate assessment of a melanoma’s invasive potential. This well help to devise faster and more effective interventions.
Clearly, AI is a technology whose full potential for dermatology is still to be discovered and tested. It must not be considered a substitute for human intelligence and medical advice care, but as a developing tool to improve diagnoses and offer more timely and effective care to patients.