Scientists have found artificial intelligence (AI) improves the accuracy of skin cancer diagnosis, when combined with human clinical checks.
In a world-first study, a team of researchers, including scientists from the University of Queensland (UQ), trialled a collaborative approach where clinicians were assisted by AI.
Professor Monika Janda from UQ’s Centre for Health Services Research said the highest diagnostic accuracy was achieved when crowd wisdom and AI predictions were combined.
“This is important because AI decision support has slowly started to infiltrate healthcare settings, and yet few studies have tested its performance in real-world settings or how clinicians interact with it,” she said.
The findings, published in Nature Medicine recently, suggest that human-AI and crowd-AI collaborations are preferable to individual experts or AI alone.
“Inexperienced evaluators gained the highest benefit from AI decision support, and expert evaluators confident in skin cancer diagnosis achieved modest or no benefit,” Janda added.
Scientists developed the AI algorithm from the publicly available HAM10000 dataset, a large collection of multi-source dermatoscopic images of common skin lesions, and used the PyTorch open source machine learning library, which was primarily developed by Facebook’s AI Research lab.
They then evaluated and quantified the performance of human raters and AI when it came to diagnosing skin cancers through the web-based platform DermaChallenge.
To do this, the researchers trained and tested an artificial convolutional neural network to analyse pigmented skin lesions, and compared the findings with human evaluations on three types of AI-based decision support.
“Our study found that good quality AI support was useful to clinicians but needed to be simple, concrete and in accordance with a given task,” Janda said.
“For clinicians of the future, this means that AI-based screening and diagnosis might soon be available to support them on a daily basis.”