Artificial intelligence has been a buzzword in medicine for some years, but it is only slowly becoming a practical reality in dermatology, with many ethical and practical considerations to iron out first.
There is substantial interest in AI in dermatology, especially for lesion detection and classification, which iToBoS aims to provide.
There have been very encouraging results with AI image classification, with several studies showing that, when dermoscopy images alone were used, algorithms could outperform experienced clinicians. However, clinicians regain the advantage when other information in included, such as patient and family history of skin cancers or the overall appearance of the patient’s other moles.
By 2020, the FDA had approved AI image classifiers for radiology, cardiology and ophthalmology, but not dermatology. A handful of image classifiers, some integrated with teledermatology, have been approved for use in the EU and Australia. There are also several direct-to-consumer apps that are outside the Therapeutic Goods Administration’s regulatory approval.
The Australasian College of Dermatology (ACD) has recently put out a position statement on the use of AI algorithms in dermatology practice for its members, focusing on appropriate use. It principally supports the development of AI models to enhance the practice of dermatology and augment care, while maintaining the ethical principles of beneficence (must benefit patients), non-maleficence (avoids harm to patients, particularly by maintaining patient privacy), and transparency (making it clear to patients that an AI model was used to inform the diagnosis.)
An important plank of the recommendations is justice and equity, in particular using AI models that attempt to reduce bias and supporting models that enhance outcomes for Aboriginal and Torres Strait Islander peoples. The ACD also encourages collaboration withy consumer groups, along with regulatory bodies and clinicians, so develop appropriate AI regulation and education for safe and effective use.
To justify widespread adoption of an AI model, the ACD advocates for prospective, real-world evaluations to show that the model enhances the performance of clinicians in the Australian context, or are least equivalent to clinicians in performance.
To maximise patient safety, the ACD recommends that only TGA-approved models be used, and that unregulated direct-to-consumer products be avoided. All the outputs from the model should be auditable and stored in the patient’s medical record.
Finally, and perhaps most importantly, the ACD recommends that AI devices only ever be used to assist a clinician in making a diagnosis, so that clinical judgement, and responsibility, ultimately remains with the clinician.
Find out more at https://onlinelibrary.wiley.com/doi/10.1111/ajd.13946