We use cookies to understand how you use our site and to improve your experience. This includes personalizing content and advertising. To learn more, click here. By continuing to use our site, you accept our use of cookies. Cookie Policy.
Features Partner Sites Information LinkXpress
Sign In
Advertise with Us
Ampronix,  Inc


28 Jan 2019 - 01 Feb 2019
02 Feb 2019 - 07 Feb 2019
16 Feb 2019 - 21 Feb 2019

AI Surpasses Dermatologists at Melanoma Recognition

By Medimaging International staff writers
Posted on 05 Jun 2018
Print article
Image: A new study claims deep-learning algorithms beat dermatologists at dermoscopy interpretation (Photo courtesy of Dreamstime).
Image: A new study claims deep-learning algorithms beat dermatologists at dermoscopy interpretation (Photo courtesy of Dreamstime).
Trained artificial Intelligence (AI) convolutional neural networks (CNNs) are better than experienced dermatologists at detecting skin cancer, claims a new study.

Researchers at the University of Heidelberg (Germany), the University of Passau (Germany), and other institutions trained the Google Inception CNN to identify skin cancer by showing it more than 100,000 stored images of malignant melanomas, as well as benign moles and nevi. They then compared the performance of the CNN with that of 58 international dermatologists via a 100-image test-set, using two levels of evidence; level-I included dermoscopy images alone, and level-II included dermoscopy plus clinical information and photos.

The dermatologists were asked to first make a diagnosis of malignant melanoma or benign mole just from the dermoscopic images (level I) and make a decision about how to manage the condition (i.e., surgery, short-term follow-up, or no action needed). Four weeks later, they were given additional clinical information on the patient--including age, sex, and position of the lesion--and close-up images of the same 100 cases (level II), and asked once again for their diagnoses and management decisions.

The results revealed that in level I, the dermatologists accurately detected an average of 86.6% of melanomas, and correctly identified an average of 71.3% of lesions that were not malignant. However, when the CNN was re-tuned to the same level as the physicians to correctly identify benign moles (71.3%), the CNN successfully detected 95% of melanomas. At level II, the dermatologists improved their performance, accurately diagnosing 88.9% of malignant melanomas and 75.7% that were not cancer. The study was published on May 28, 2018, in Annals of Oncology.

“The CNN missed fewer melanomas, meaning it had a higher sensitivity than the dermatologists, and it misdiagnosed fewer benign moles as malignant melanoma, which means it had a higher specificity; this would result in less unnecessary surgery,” said lead author Professor Holger Haenssle, MD, of the University of Heidelberg. “When dermatologists received more clinical information and images at level II, their diagnostic performance improved. However, the CNN, which was still working solely from the dermoscopic images with no additional clinical information, continued to out-perform the physicians' diagnostic abilities.”

“This CNN may serve physicians involved in skin cancer screening as an aid in their decision whether to biopsy a lesion or not. Most dermatologists already use digital dermoscopy systems to image and store lesions for documentation and follow-up,” concluded Professor Haenssle. “The CNN can then easily and rapidly evaluate the stored image for an 'expert opinion' on the probability of melanoma. We are currently planning prospective studies to assess the real-life impact of the CNN for physicians and patients.”

Deep learning is part of a broader family of AI machine learning methods based on learning data representations, as opposed to task specific algorithms. It involves neural network algorithms that use a cascade of many layers of nonlinear processing units for feature extraction and transformation, with each successive layer using the output from the previous layer as input, thus forming a hierarchical representation.

Related Links:
University of Heidelberg
University of Passau

Print article


Copyright © 2000-2019 Globetech Media. All rights reserved.