26 July 2023

The AI ‘avalanche’ is coming

Clinical Conferences Technology

A concurrent session at the recent ACD conference explored patient, clinician and researcher views on the use of AI in dermatology.

Patients, clinicians and researchers all support the use of AI in dermatology, but that doesn’t mean there aren’t concerns over how it should be – and is – used.

Speaking to delegates at the Australasian College of Dermatology Annual Scientific Meeting in Sydney earlier this year, a series of speakers discussed the implications of using AI in clinical practice, presenting views consumers, clinicians and researchers on the topic.

The real consumer of AI may not be who you think it is

Mr Craig Lawn, a consumer advisor to the Australian Centre of Excellence in Melanoma Imaging and Diagnosis, began his presentation by discussing the impending AI “avalanche”.

“We cannot afford the time to sit and ponder or be conservative. There are billions of dollars being thrown at [AI] in Silicon Valley now.”

Mr Lawn encouraged dermatologists to play an active role in “setting the system at the right levels” to ensure AI is used in an appropriate and sustainable way moving forward, regardless of whether you use or trust the technology.

“The spoiler is it’s not me [who is the consumer of AI]. You are the consumers of AI,” Mr Lawn explained.

“Healthcare is incredibly complex, and we absolutely trust you guys. [But] do you trust and recommend the technology, software and applications driven and enhanced by AI?” Mr Lawn asked.

While patients often share similar concerns as their doctors regarding the use of AI in healthcare (e.g., the potential for bias, concerns over diagnostic accuracy, etc.), Mr Lawn encouraged attendees to consider the potential benefits AI could bring. 

Mr Lawn also stressed the importance of maintaining good relationships between patients and their doctors in the face of AI becoming increasingly common in clinical care.

“We don’t want to be an image or a bit of data,” he told delegates, “[but] there is overwhelmingly strong support for a ‘doctor plus’ approach from consumers,” where AI is used to support and enhance clinical decision-making, rather than replace it.

The College also recommended a “doctor plus” approach in their position statement on the use of AI in dermatology in Australia, released earlier this year.

You can’t believe everything you hear from AI

Associate Professor Victoria Mar, director of the Victorian Melanoma Service at The Alfred in Melbourne, also spoke about the importance of dermatologists playing an active role in the development and implementation of AI-based technologies.

“We know that AI models sometimes struggle with [diagnosing] lesions that we might not struggle with. And the same lesions we find difficult to diagnose are not necessarily the same lesions the computer would find difficult to diagnose,” she said.

Although several observational studies use in silico tests to compare the performance of AI algorithms against that of dermatologists, there is a paucity of clinical trials exploring how AI performs in a more realistic clinical setting, especially when patients are involved.

To address this, Professor Mar and colleagues undertook a pre-post intervention trial of an AI system used as a diagnostic aid for skin cancer management, publishing their results online late last year.

Using an AI-assisted diagnosis tool helped registrars select management plans that were more aligned with the consultants, meaning patients were treated in a more consistent fashion.

However, several patients insisted on having their lesion biopsied when the AI tool was unsure or called the lesion malignant – even if the dermatologist said it was benign.

Other studies performed overseas also reported similar experiences, where patients or doctors second-guessing their judgement after consulting an AI tool resulted in unnecessary biopsies of benign lesions.

“I know Craig thinks patients trust [doctors], but they also really trust the AI. [So] we need to be really vigilant in testing these sorts of algorithms in a clinical setting,” Professor Mar concluded.

Testing the mood of the electorate

Delegates completed a short poll about how they view and use AI in dermatological practice following presentations from Mr Lawn and Professor Mar, with a real time discussion of the results.

While only nine of the 130-odd respondents indicated they currently use AI in their practice, over 80 respondents were planning to start using it in the future.

At least 90 respondents felt AI could be helpful for clinicians, with roughly even numbers indicating they would consider using AI to obtain a second opinion as well as to help triage patients.

Despite this apparent support for the use of AI in clinical practice, delegates expressed concerns on how the use of the technology would be regulated.