Artificial intelligence (AI) detects melanoma more efficiently than dermatologists
By Eric Backus, MBA (MSOC Client Manager)
What’s the question?
A recent study published in the Annals of Oncology by the Oxford University Press aimed to identify whether artificial intelligence (AI) was able to better detect melanoma than trained dermatologists. This study compared a convolutional neural network (CNN) to 58 dermatologists and their ability to recognize dermoscopic melanoma. CNN is primarily used to analyze images and visual recognition through a combination of algorithms and AI hardware that can learn by example. The AI learned to recognize skin cancer by examining over 100,000 different images of benign and malignant melanomas/moles.
What did the researchers find?
This study compared the performance of 58 dermatologists across 17 countries to the CNN technology. The dermatologists’ experience ranged from beginners with less than two years to experts with more than five years. Most of these dermatologists are considered experts in dermoscopy. To compare their abilities, both the CNN and the dermatologists analyzed 100 images of skin lesions, decided whether the lesion was benign or malignant, and stated what the next step in treatment should be. The three treatment options included surgery, follow-up visit, or no further action. After four weeks, the dermatologists were provided with additional information, including further information about the patient and different close-up images.
At the conclusion of this study, it was shown that the AI properly identified malignant melanoma 95% of the time, while dermatologists were correct 86.6% of the time without additional information and 88.9% of the time with the additional information. The CNN had more difficulty identifying benign moles at just 63.8% correct while dermatologists were at 71.3% without additional information and 75.7% with additional information. However, when the CNN was given more images of benign moles to analyze, its percentage increased to 80%. Therefore, the investigators concluded that the AI was able to more efficiently and accurately identify malignant melanoma than dermatologists.
So what’s the takeaway?
Despite the promising findings of this study, it would not be wise to immediately replace all dermatologists with AI. Dermatologists have a large variety of expertise that could not be replicated by AI alone. Patients, especially those with possible melanoma, seek an office-based, in-person experience with a doctor. While this study shows strong support that AI can help determine the difference between melanoma and benign moles, the best outcomes and most accurate identification would likely occur when the expertise of dermatologists is combined with the efficiency of AI analysis.
One potential use of this AI would be through a smartphone app. Over six billion smartphones are projected to be in use by 2021. With the increasing quality of smartphone cameras, CNN could use photos taken by patients at home to help identify skin cancer. This could drastically cut down on time spent in the office and provide dermatologists with an initial diagnosis and treatment plan before the patient is even seen. An app-based skin cancer screening also has the potential to help patients across the world who don’t have easy access to a dermatologist. The evolution of AI in the medical field is constantly improving, so it would certainly be no surprise to see this CNN technology used in a medical practice near you.