Deep Learning Algorithm Does as Well as Dermatologists in Identifying Skin Cancer




It is really a terrible feeling to get your strange moles tested at health care center to identify if it is cancerous. The situation becomes more troublesome when you have to travel so far to meet an experienced doctor to deal with your issue. Considering this problem, we must say that there is a great need of developing some life-saving way to avail diagnosis on a smartphone.
With the same thought, computer scientists at Stanford made an effort to develop an Artificial Intelligence-based algorithm for diagnosing skin cancer. They worked on a database of 130,000 images collected from people suffering from skin disease.

Note that, every year almost 5.4 million new cases of skin cancer are reported in the United States out of which only 14% are detected on early stage and rest all lose the hope of five-year survival rate as well. Hence, it is important to develop algorithms for early detection of skin cancer. When the diagnosis is carried by dermatologists they start with a visual inspection that has a lower rate of accuracy. On the other side, latest Artificial Intelligence-based models are able to detect the cancerous cells on time.

Instead of working on a new algorithm, researchers decided to modify the already developed algorithm of Google. They have trained it for differentiating between dogs and cats by using 1.28 million images. Now as it is not so easy to find such a huge dataset for skin cancer patients so researchers decided to take 130,000 images of skin lesions that were related to 2000 different diseases. However, they finally used only 370 high-quality images of cancerous and noncancerous lesions that were identified by dermatologists.
Performance of the training algorithm was tested on the basis of the sensitivity-specificity curve that represented 91% of the total area of the graph. One of the biggest benefits of this algorithm is that its response can be adjusted for different sensitivity levels depending upon relevant details that person wants to extract.

Currently, this algorithm is available on the computer but the team is working hard to make it compatible with a smartphone so that patients can have a reliable solution for cancer diagnosis at their fingertips. It is believed that this latest technology will assist dermatologists to diagnose disease symptoms at an early stage so that appropriate treatment can be applied on time.