Skin cancer is the most common kind of cancer and quite often its early detection is crucial for the successful subsequent treatment. Unfortunately, not all people have easy access to a dermatologist for one reason or another. Therefore, scientists have been working on the development of an alternative method, which should combine both accuracy, reliability and accessibility.
In its issue from 25th January Nature has reported that computer scientist from Stanford University have developed an algorithm that can detect and recognize skin cancer, which was cross checked against 21 dermatologists and performed as good as them.
The first test to determine the presence of skin cancer is visual. Doctors examine the skin by looking at the mole or lesion of concern using their naked eye or a dermatoscope – a small handheld microscope. If there is any doubt or concern then biopsy is the next step.
The development of an algorithm that can make automated classification of skin lesions using images was started as a class project by professor Sebastian Thrun working at the Stanford Artificial Intelligence Laboratory and soon proved to be much more. Thrun and his team developed a deep learning algorithm that incorporates visual processing to identify types of skin cancer. Thus they had a type of artificial intelligence that follows the model of neural networks in the brain. The idea of deep learning is not a new concept in computer science and has been used for many decades but only recently it included visual processing tasks. The scientist did not "invent" the algorithm themselves but started their work based on an algorithm developed by Google that was trained to identify more than a million images from 1,000 object categories. In their case, however, it had to do something more than differentiate a cat from a dog, namely to identify malignant carcinoma form benign seborrheic keratosis.
In order to do that, the scientists first had to create a database of clinical images of different diseases, which to be fed to the algorithm. After a lot of work, translation and collection of data, they managed to arrange a dataset of nearly 130,000 images of skin lesions from 2,032 different diseases.
“We made a very powerful machine learning algorithm that learns from data,” commented Andre Esteva, co-lead author of the paper published in Nature, who worked closely with Thrun on the project.
The algorithm that uses pixels and disease labels as inputs was tested against 21 dermatologists over more than 370 biopsy-proven clinical images. The task was to check its sensitivity and specificity by showing ability to correctly identify both benign and malignant lesions or in other words to be able to spot the most common cancers and the deadliest type of skin cancer. The algorithm achieved the same results as the doctors showing a level of competence equal to certified dermatologists.
So far, the algorithm is only computer-based, however the scientists believe that it could be transferred to mobile devices and that soon anyone with a smartphone in his hand will be able to check a suspicious skin lesion. The algorithm, of course, should be validated before being implemented in practice by both dermatologists and patients. Nevertheless, scientists believe that deep learning could be beneficial for visual diagnosis of other diseases or in other medical fields in near future.
Meantime, while waiting for a handheld device that can help us identify the type of lesion, we shall take proper care of our skin. Applying natural and organic products is a step towards protecting our skin from adverse weather conditions. Pure argan oil is an excellent way to revitalize your skin, hair and nails. The argan oil is recognized worldwide for its benefits both for the human health and the environment and some research has even shown that it can have positive anticancer effect. Take good care of your body and skin in particular and reap the positive results of your investment.