Apr 1 2020
Korean scientists have created a deep learning-based artificial intelligence (AI) algorithm that can precisely categorize cutaneous skin disorders, estimate the probability of malignancy, propose primary therapeutic options, and act as a supplementary tool to improve clinicians’ diagnostic accuracy.
Examples of output from Model Dermatology (http://modelderm.com), showing the top-three choices for each skin lesion. Left: a case of basal cell carcinoma that is commonly misdiagnosed as a nevus. Right: a case of eczema herpeticum that is commonly misdiagnosed as atopic dermatitis. In both cases, the authors’ algorithm correctly diagnosed the condition (top choice). Image Credit: Model Dermatology.
Using this AI system, the accuracy of diagnoses made by dermatologists and also by the general public was considerably enhanced. This innovative study has been reported in the Journal of Investigative Dermatology.
While skin diseases are common, it is not always simple to differentiate malignant from benign conditions, or to see a dermatologist immediately.
Recently, there have been remarkable advances in the use of AI in medicine. For specific problems, such as distinguishing between melanoma and nevi, AI has shown results comparable to those of human dermatologists.
Jung-Im Na, MD, PhD, Study Lead Investigator, Department of Dermatology, Seoul National University, Seoul, Korea
Dr Na continued, “However, for these systems to be practically useful, their performance needs to be tested in an environment similar to real practice, which requires not only classifying malignant versus benign lesion, but also distinguishing skin cancer from numerous other skin disorders including inflammatory and infectious conditions.”
The researchers employed a unique AI algorithm called “convolutional neural network” and ultimately developed an AI system that has the ability to classify skin disorders, predict malignancy, and suggest treatment options.
The researchers collected as many as 220,000 images of Caucasians and Asians who were suffering from 174 skin diseases and taught neural networks to decode those images.
The team observed that the AI algorithm not only diagnosed 134 skin disorders and proposed primary treatment options, but it also rendered multi-class classification among the various skin disorders and improved medical professionals’ performance via augmented intelligence. A majority of earlier researches had been confined to particular tasks, like distinguishing melanoma from nevi.
The performance of the AI algorithm was originally compared with the performance of 26 resident dermatologists, 21 dermatologists, and 23 members of the general public. It was observed that the performance of the AI algorithm was analogous to that of the dermatology residents but marginally less than that of the dermatologists.
Post the preliminary test, the algorithm results were revealed to the test participants and their answers were subsequently altered.
The sensitivity of the diagnosis of malignancy made by 47 clinicians enhanced from 77.4% to 86.8%. Likewise, the sensitivity of the malignancy diagnosis made by the 23 members of the general public considerably improved from 47.6% to 87.5%.
Most importantly, based on the preliminary results, 50% of the malignancies were probably overlooked by the general public without referral to medical specialists.
Our results suggest that our algorithm may serve as an Augmented Intelligence that can empower medical professionals in diagnostic dermatology. Rather than AI replacing humans, we expect AI to support humans as Augmented Intelligence to reach diagnoses faster and more accurately.
Jung-Im Na, MD, PhD, Study Lead Investigator, Department of Dermatology, Seoul National University, Seoul, Korea
The scientists, however, warned that the AI algorithm cannot conclusively decode the images—especially the images that it has not trained to understand even when the problem presented was very simple. For instance, an algorithm that is trained solely to distinguish melanoma from nevi cannot distinguish between a picture of a nail hematoma and a nevus or a melanoma.
If the hematoma has an irregular shape, the AI algorithm is likely to diagnose it as melanoma. The researchers also pointed out that the AI algorithm was tested and trained utilizing high-quality images and, if the input images have low quality, its performance would be usually substandard.
Furthermore, a diagnosis made with just a single image with the best composition may pose inherent restrictions when compared to the diagnoses made in a clinical environment.
In actual practice, the diagnosis made by a dermatologist is based on a mix of numerous sources of data such as symptoms, past medical history, the texture of the lesion reviewed by physical contact, and its appearance when compared to the other lesions on the patient.
We anticipate that the use of our algorithm with a smartphone could encourage the public to visit specialists for cancerous lesions such as melanoma that might have been neglected otherwise. However, there are issues with the quality or composition of photographs taken by the general public that may affect the results of the algorithm.
Jung-Im Na, MD, PhD, Study Lead Investigator, Department of Dermatology, Seoul National University, Seoul, Korea
Dr Na continued, “If the algorithm’s performance can be reproduced in the clinical setting, it will be promising for the early detection of skin cancer with a smartphone. We hope that future studies will evaluate the utility and performance of our algorithms in a clinical setting.”
A previous demo version of the team’s deep learning method is now available through its website. By examining data via this website, the team is hoping to detect potential problems that may emerge if the AI algorithm was used through telemedicine, which largely depends on clinical photography to identify skin disorders.
But diagnoses like these will still have to be validated by dermatologists along with the patient’s physical examination and medical history.
Dr Na and Sung Eun Chang, MD, Department of Dermatology, Asan Medical Center, Ulsan University College of Medicine, Seoul, Korea, had contributed equally and critically to this research.
Moreover, primary scientists in the team included Seung Seog Han from I Dermatology Clinic, and Ilwoo Park from Chonnam National University.