Machine learning outperforms human dermatologists in diagnosing pigmented skin lesions

State-of-the-art machine learning algorithms diagnose benign and malignant skin lesions more accurately than dermatologists, dermatology residents and general practitioners, according to a study published June 11 in The Lancet Oncology.

Researchers compared the diagnostic abilities of a group of 511 healthcare professionals to that of 139 algorithms, each trained on a set of more than 10,000 dermatoscopic images and initially developed for the International Skin Imaging Collaboration challenge in 2018. In the study, the humans and algorithms were each asked to diagnose a set of 30 images, selected randomly from a 1,511-image test set.

On average, the algorithms achieved about two more correct diagnoses per 30-image batch than the entire human group. Furthermore, the top three algorithms made an average of 25.43 correct diagnoses per set, while 27 human experts, each of whom had at least 10 years of experience, achieved fewer than 19 correct answers out of every set of 30 images.

Though the artificial intelligence proved slightly less adept at diagnosing types of images that had not been in the training set, the study's authors maintained that, overall, the results prove that these machine learning algorithms "should have a more important role in clinical practice."

More articles about AI:
Humana, Peninsula Regional leaders make IBM's 'Women Leaders in AI' list
FDA clears AI tool for triage of cervical spine fractures
U of Virginia physicians work with data scientists to develop predictive model for dosing protocols

© Copyright ASC COMMUNICATIONS 2020. Interested in LINKING to or REPRINTING this content? View our policies by clicking here.


Featured Content

Featured Webinars

Featured Whitepapers