Study: AI tool detects skin cancers better than dermatologists

An AI-based network outperformed several dermatologists while analyzing skin lesions for cancer, suggesting the technology could be a helpful tool for physicians regardless of their experience, according to a recent study.  

“Our data clearly show that a CNN (convolutional neural network) algorithm may be a suitable tool to aid physicians in melanoma detection irrespective of their individual level of experience and training,” study author Holger A. Haenssle, MD, a professor with the department of dermatology at the Heidelberg University in Germany, et al. wrote.

The study was recently published in the Annals of Oncology.

For the study, researchers trained and tested a deep-learning CNN on how well it could diagnosis skin lesions with melanoma or as benign nevi based on a test set of 100 images. They then compared the CNN's performance to 58 dermatologists, including 30 experts from 17 different countries.

The CNN “significantly outperformed” the dermatologists when analyzing the dermoscopic (skin lesion) images only. When using the dermatologists’ mean sensitivity of 86.9 percent, the CNN’s specificity was higher, 82.5 percent, while the dermatologists’ specificity was 71.3 percent.

Dermatologists performed better after adding real-life clinical information to their diagnoses, but was still outperformed by the CNN, according to researchers. When using the dermatologists’ improved mean sensitivity of 88.9 percent, the CNN had a specificity of 82.5 percent, while they reported 75.7 percent.

“In conclusion, the results of our study demonstrate that an adequately trained deep learning CNN is capable of a highly accurate diagnostic classification of dermoscopic images of melanocytic origin,” the authors concluded. “In conjunction with results from the reader study (levels), we could show that the CNN’s diagnostic performance was superior to most but not all dermatologists.”

Based on the results, researchers suggested physicians of “all different levels of training and experience may benefit from assistance by a CNN’s image classification.”