Novel AI imaging approach yields improved skin cancer diagnosis

A novel imaging approach using artificial intelligence was associated with improved detection of parameters associated with melanoma, according to results presented at the International Conference on Image Analysis and Recognition.

The researchers suggested that a number of quantitative imaging approaches to dealing with melanoma have focused largely on skin lesions using “hand crafted imaging features.” The current study employed “machine-learning” software, which records abstract quantitative features on images and can model physiological traits of the patients, according to study background.

The two features of melanoma that were assessed in the analysis were eumelanin and hemoglobin concentrations observed on dermal imaging. The researchers created a non-linear random forest regression model culled from the images.

Results indicated that the novel approach was superior to current techniques in terms of predicting eumelanin and hemoglobin. The findings were validated by a test that was applied to the clinical images.

In an interview with Healio.com/Dermatology, Alexander Wong, PhD, associate professor, department of systems design engineering at the University of Waterloo, Ontario, Canada, said the take-home message of the study is that melanoma, while on the rise worldwide and highly deadly if detected too late, is highly treatable if caught early.

“Artificial intelligence [AI] can be a key ingredient in the battle against not just melanoma, but also skin cancer globally,” he said.

Wong added that these findings show that AI can be a very powerful tool to enhance screening methods by providing clinicians with insightful physiological information. This will allow them “to not only more readily diagnose and treat skin cancer, such as melanoma, more accurately and consistently, but also do it at a much earlier stage by enabling general practitioners and nurse practitioners ... to better screen for skin cancer before they reach specialized dermatologists and biopsies, thus reducing health care costs as well as wait times to diagnosis,” he said. “Our goal is to get portable, handheld imaging devices powered for such artificial intelligence ready and accessible to clinicians worldwide.” – by Rob Volansky

Reference:

Cho DS, et al. A machine learning-driven approach to computational physiological modeling of skin cancer. Presented at: The International Conference on Image Analysis and Recognition; July 5-7, 2017; Montreal.

Disclosure s : Wong was a co-developer of the imaging technology evaluated in this study. Healio.com/Dermatology was unable to obtain relevant financial disclosures for other authors at the time of publication.