Algorithm Can Diagnose Skin Cancer: Deep Learning Process Makes The Breakthrough

• comments
• 
• 

It is surprising but true that algorithm can identify skin cancer. Yes, computer scientists at Stanford University have created one such thing to diagnose skin cancer. This diagnosis truly creates a new dimension in the arena of medical science.

Stanford News reports that the scientists of the Stanford University created a database of 130,000 clinical images of skin disease. This database helps the algorithm to diagnose skin cancer. Sebastian Thrun, the professor in the Stanford Artificial Intelligence Laboratory, said that they tried to invent something that can work like the dermatologist.

Most of the times, dermatologists use the dermatoscope to check lesion to identify skin cancer. If the dermatologist believes that the lesion is cancerous, then a biopsy test is the ultimate step to complete the diagnosis. The new artificially diagnosis algorithm process can enhance this diagnosis process accurately.

The algorithm examination process is the combination of the visual process with deep learning. Deep learning is an essential process in computer science to figure out a problem. Currently, scientists use it to perform visual processing tasks. Nature reports that the final product was tested against 21 board-certified dermatologists and the result was inspiring.

 Andre Esteva, co-lead author of the research paper and also a graduate student in the Thrun lab, said that they have invented the powerful machine that can understand the algorithm. The process depends on data. The author also uttered that it can figure out the actually required fact.

The unique thing is the researchers started their work with an algorithm developed by Google. The key fact is it can already identify 1.28 million images from the 1,000 object categories. The number of the images indicates the huge capability of this process.

Brett Kuprel, co-lead author of the research paper and a graduate student in the Thrun lab, said that they collected images from the internet to create a dataset of skin cancer. They created this dataset to train the algorithm. The researchers worked together with the dermatologists at Stanford Medicine.

Helen M. Balu, professor of microbiology and immunology at Stanford University joined the research team to collect images from the internet. They together collected 130,000 images to create the ultimate algorithm. All the images belong to skin lesions that represent more than 2,000 different diseases.

The researchers tested and assessed the performance of algorithm through a "sensitivity-specificity curve". Three key tasks helped to measure the performance. The result was very satisfactory.

The best advantage of the algorithm is it can respond to the required assessment. That means it can respond to less or more sensitive requirements. The whole process can properly evaluate the provided skin lesion images.

Currently, the algorithm exists on a computer. The good news is the research team is trying to make it smartphone compatible. The success of this effort will surely help to diagnose the skin cancer more easily and effectively.