Artificial Intelligence (AI) Performs Better Than Dermatologists in Diagnosing Skin Cancer

Machine learning or artificial intelligence that is fed over timewith thousands of images of benign skin moles (nevi) and melanoma (skin cancer) and trained to recognize even the minute differences between the two conditions has been found to perform better than experienced dermatologists in the correct diagnosis of melanoma. The study was conducted by a research team from Germany, the USA and France who trained the machine or CNN. The findings of their study appear in the cancer journal Annals of Oncology.

Advertisement

Details of Study

• To train the machine, the team showed the machine more than 100,000 images of previously diagnosed malignant and benign skin cancers and moles. Only dermoscopic images were used, i.e. lesions that were imaged at a 10-fold magnification.
• With each training session and more images shown, the CNN improved its ability to distinguish between benign and malignant lesions.

The first author of the study, Professor Holger Haenssle, senior managing physician at the Department of Dermatology, University of Heidelberg, Germany, explained: "The CNN works like the brain of a child.”

• Once the training period finished, the team created two test sets of images from the Heidelberg library that had never been used for training and therefore not known to the CNN. One set of 300 new images was built to solely analyze the accuracy of the CNN. Before that, 100 of the most difficult lesions were selected to test actual dermatologists in comparison to the results of the CNN.
• Dermatologists all over the world were invited to take part, and 58 skin specialists from 17 countries agreed. Among these, 17 (29%) stated they had less than two years' experience in dermoscopy, 11 (19%) said they had between two to five years' experience, and 30 (52%) were deemed expert with more than five years' experience.
• The dermatologists had to first make a diagnosis of benign mole or malignant melanoma only from the dermoscopic images (level I) and give opinion about management (surgery, short-term follow-up, or no action needed). Four weeks later they were given more details such as clinical history about the patient such as age, sex and site of the lesion with close-up images of the same 100 cases (level II) and had to again offer a diagnosis and management plan.
• In level I, the dermatologists correctly diagnosed an average of 86.6% of melanomas, and an average of 71.3% of lesions that were benign. However, for the same level, the CNN detected 95% of melanomas and same percentage (71.3%) of benign moles. At level II, the dermatologists improved their performance, correctly identifying 88.9% of malignant melanomas and 75.7% that were not cancer.

According to Professor Haenssle, "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."

• The expert dermatologists did better at level I than the less experienced dermatologists and were able to better identify malignant melanomas. However, their ability to make the correct diagnosis on average was still worse than the CNN at both levels I and II.

Thus, the results of the study suggest that the CNN outperformed the dermatologists accuracy with a high degree of sensitivity as well as specificity including highly experienced dermatologists to diagnose melanomas.

Advertisement

What is Artificial Intelligence or Deep Machine Learning/CNN?

A CNN is an artificial neural network similar to nerve cells (neurons) network in the brain that are connected to each other and respond to what the eye sees. The CNN is capable of learning rather quickly from images that it "sees" and training itself over time with several layers of input and to learn to gradually improve its performance (a process known as machine learning) and give the desired output.

Advertisement

Scope of Study and Future Plans

• The authors of the study believe that it may one day be possible for artificial intelligence to help make accurate diagnosis of skin cancer early with much better chances of a cure for patients.

Professor Haenssle said: "I have been involved in research projects that aim at improving the early detection of melanoma in its curable stages for almost 20 years. My group and I are focusing on non-invasive technologies that may help physicians not to miss melanomas, for instance, while performing skin cancer screenings. When I came across recent reports on deep-learning algorithms that outperform human experts in specific tasks, I immediately knew that we had to explore these artificial intelligence algorithms for diagnosing melanoma."

• However, they do not think that AI can take over from doctors in making a diagnosis of skin cancer but would prove to be an additional valuable aid to correct diagnosis.
• The CNN could help doctors involved in skin cancer screening to decide if they need to biopsy a given lesion or not.
• Most dermatologists already use digital dermoscopic devices to image and store lesions for documentation and future follow-up. The AI can then thus easily and rapidly evaluate the stored image for an 'expert opinion' on presence or absence of skin cancer.
• They team are also planning prospective studies to test the real-life impact of the CNN for both physicians and patients.

Potential Limitations of Study

The study has certain drawbacks, such as:

• The dermatologists were in an experimental setting, knowing they were not making "life or death" decisions.
• The test images did not include the entire range of skin lesions.
• There were less number of images from non-Caucasian skin types and genetic background.
• The fact that doctors may be reluctant to follow the recommendation of a CNN they are unfamiliar with.
• Before using AI as standard procedure in clinics, the authors feel that certain issues such as difficulty of imaging melanomas on certain sites such as the fingers, toes and scalp, and training AI sufficiently to recognise atypical melanomas have to be first addressed.

In conclusion, "Currently, there is no substitute for a thorough clinical examination. However, 2D and 3D total body photography is able to capture about 90 to 95% of the skin surface and given exponential development of imaging technology we envisage that sooner than later, automated diagnosis will change the diagnostic paradigm in dermatology. Still, there is much more work to be done to implement this exciting technology safely into routine clinical care."

References:

1. V J Mar, H P Soyer "Artificial intelligence for melanoma diagnosis: How can we deliver on the promise?" (2018) Annals of Oncology https://doi.org/10.1093/annonc/mdy193
2. Tests for Melanoma Skin Cancer - (https://www.cancer.org/cancer/melanoma-skin-cancer/detection-diagnosis-staging/how-diagnosed.html)
3. Skin Cancer - Tests to diagnose - (http://www.cancerresearchuk.org/about-cancer/skin-cancer/getting-diagnosed/tests-diagnose)
4. Skin Cancer - (https://www.mayoclinic.org/diseases-conditions/skin-cancer/symptoms-causes/syc-20377605)

Source-Medindia