A direct-to-consumer machine learning model for detecting skin cancers available on app stores incorrectly classified rare and aggressive cancers as low-risk, a conference has heard.
Following the findings, which were presented at the European dermatology event, the EADV Congress, yesterday, the research team said it was ethically questionable to produce apps without any transparency on performance metrics for rare but potentially life-threatening skin cancers.
Researchers from Queen Mary University of London and The Royal London Hospital focused on Merkel cell carcinoma (MCC) and amelanotic melanoma, both are aggressive cancers that grow quickly and need early treatment.
They created a dataset of 116 images of these two rare cancers and of benign lesions seborrhoeic keratosis and haemangiomas, and assessed the images with two machine-learning models.
The first model studied was a certified as a medical device, was directly sold to the public via the App store and was advertised as being able to diagnose 95% of skin cancers. The second model was available for research purposes only and used as a reference.
The researchers found that Model 1 incorrectly classified 17.9% of MCCs and 22.9% of amelanotic melanomas as low-risk, while classifying 62.2% of benign lesions as high risk.
For detecting malignancy, Model 1’s sensitivity was 79.4% – with a 95% confidence interval (CI) 69.3-89.4% – and specificity was 37.7% – 95% CI 24.7-50.8.
When examining Model 2, MCC was not included in the top five diagnosis for any of the 28 MCC images analysed, which, say the researchers, raises the possibility that the model had not been trained to recognise the disease.
They warn that Model 1’s high false positive rate has potentially negative consequences, while the results pose a bigger question of the safety of other artificial intelligence (AI) models for detecting skin cancer that are available on the market.
Lead author Lloyd Steele, of the Blizard Institute, Queen Mary University of London, said: “In order to improve, machine learning model evaluations should consider the spectrum of diseases that will be seen in practice. At the moment, most of the performance of those models is driven by the imaging data available, which is particularly scarce when it comes to rare skin cancers.”
Marie-Aleth Richard, EADV board member and professor at the University Hospital of La Timone, Marseille, added: “The number of skin cancer detection apps available for consumer use is growing, but as demonstrated in this research, there must be more transparency around the safety and efficacy of these apps.
“Furthermore, such devices detect only what they are shown to analyse and do not make systematic analysis of all the skin’s surface. Failure to be transparent could put lives at risk.”
Steele L, Velazquez-Pimentel D, Thomas RB. Do AI Models recognise rare, aggressive skin cancers? An assessment of a direct-to-consumer app in the diagnosis of Merkel Cell Carcinoma and amelanotic carcinoma, submitted to EADV 30th Congress.
Leave a Reply