- Artificial intelligence (AI) using
deep learning convolutional neural network (CNN) shown to be more precise
in the diagnosis of melanomas compared to experienced dermatologists.
- Melanoma is often diagnosed only
late when it is more difficult to treat; early diagnosis improves chances
of a cure with favorable prognosis for the patient.
- The incidence of malignant melanoma
is rising with about 232,000 new cases globally and around 55,500 deaths
from the disease every year.
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.
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."
‘Artificial intelligence missed fewer melanomas (i.e. higher sensitivity) and misdiagnosed fewer benign moles as malignant melanoma (i.e. higher specificity) compared to dermatologists.’
- 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.
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'
- 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
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.
What is Artificial
Intelligence or Deep Machine Learning/CNN?
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
"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.
Scope of Study and
- 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.
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
- They team are also planning
prospective studies to test the real-life impact of the CNN for both
physicians and patients.
The study has certain drawbacks, such as:
- The dermatologists were in an
experimental setting, knowing they were not making "life or
- 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.
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
- 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
- Tests for Melanoma Skin Cancer - (https://www.cancer.org/cancer/melanoma-skin-cancer/detection-diagnosis-staging/how-diagnosed.html)
- Skin Cancer - Tests to diagnose - (http://www.cancerresearchuk.org/about-cancer/skin-cancer/getting-diagnosed/tests-diagnose)
- Skin Cancer - (https://www.mayoclinic.org/diseases-conditions/skin-cancer/symptoms-causes/syc-20377605)