Convolutional Neural Networks Increase Dermatologists’ Skin Cancer Diagnostic Accuracy

Key points to remember:

• Specificity and accuracy rates were higher among convolutional neural networks (CNNs).
• Dermatologists with less than 5 years of experience have seen increased diagnostic accuracy when reviewing CNN results.

Integrating convolutional neural networks into dermatology practices, especially those with less than 5 years of experience, may increase the diagnostic accuracy of skin cancer, according to a prospective study.

“In skin cancer classification tasks, convolutional neural networks (CNN) achieves diagnostic accuracies similar to those of trained dermatologists,” Julia K. Winkler, MD, from the Department of Dermatology at the University of Heidelberg in Germany, and colleagues wrote. “With the present study, we aim to elucidate dermatologist cooperation with a market-approved CNN in a prospective clinical setting.”

In this prospective diagnostic study, 22 dermatologists with varying levels of experience using dermoscopy to skin cancer screenings performed full body examinations on 188 patients. The dermatologists then indicated the level of malignancy of the suspicious lesions and proposed a management plan.

After this examination, the patients were sent to a separate room for a CNN evaluation. These results were passed on to dermatologists who were then asked to reassess their decisions based on the new findings.

The results showed that the dermatologists detected 228 suspicious melanocytic lesions, including 190 nevi and 38 melanomas.

When assessed separately, dermatologist results were comparable to CNN in terms of sensitivity (84.2%; 95% CI, 69.6% to 92.6% vs. 81.6%; 95% CI , 66.6% to 90.8%); however, CNN outperformed dermatologists in specificity (72.1%; 95% CI, 65.3% to 78% vs. 88.9%; 95% CI, 83.7% to 92.7%) and precision (74.1%; 95% CI, 68.1%-79.4% versus 87.7%; 95% CI, 82.8%-91.4%).

More importantly, when dermatologists incorporated CNN results into their decision-making, sensitivity increased to 100% (95% CI, 90.8% to 100%), specificity increased to 83.7% (95% CI, 77.8%-88.3%) and accuracy increased to 86.4% (95% CI, 81.3%-90.3%).

Additionally, with the inclusion of CNN results, unnecessary excisions of benign nevi were reduced by 19.2% (P < .001).

The authors noted that the addition of CNN results primarily benefited dermatologists with less than 5 years of experience. The results showed that the diagnostic accuracy of dermatologists with less than 2 years of experience increased from 70.5% to 87.2% once the CNN results were included (P < .01).

Dermatologists with less than 5 years of experience followed a similar pattern (77.1% vs. 91.7%; P < 0.01) whereas dermatologists with more than 5 years of experience saw no statistical increase with the addition of CNN results (74.1% vs. 75.9%).

“Dermatologists significantly improved their diagnostic performance by cooperating with the tested CNN,” the authors concluded. “These results indicate that wider application of this human-machine approach, particularly in non-specialized institutions, could be beneficial for clinicians and patients.”