A new study has found that machines using deep learning convolutional neural networks (CNN) outperformed dermatologists at detecting skin cancer.
The University of Heidelberg researchers compared a CNN’s diagnostic performance with a large international group of 58 dermatologists, including 30 experts.
For the study, Google’s Inception v4 CNN architecture was trained and validated using dermoscopic images and corresponding diagnoses for melanoma detection.
When the dermatologists were given a photo of the skin to analyse, they accurately identified 87% of the melanomas.
The deep learning convolutional neural networks performed better, with an accuracy of 95%.
“Most dermatologists were outperformed by the CNN. Irrespective of any physicians’ experience, they may benefit from assistance by a CNN’s image classification,” stated the study.