AI diagnoses skin cancer with an accuracy rate of 91%: "mobile phone" alternative doctor is not a dream

If one day, you suddenly find that a cockroach on your body becomes a bit strange, what do you do? Although this may be a dangerous signal, many people often do not check in time because of busy work, inconvenience to the hospital, and so on. Now, artificial intelligence provides a better solution to this problem: in the future, we may be able to download an app on the phone, open a camera and let the machine doctor help us to see if this is an early symptom of skin cancer.

A joint research team at Stanford University developed an artificial intelligence for the diagnosis of skin cancer that is comparable to that of human doctors. The results were published for the end of January in the cover of Nature, entitled "Dermatological Screening for Dermatologists" Dermatologist-level classification of skin cancer with deep neural networks. They used deep learning methods to identify the skin cancer symptoms with nearly 130,000 images of sputum, rash and other skin lesions. After comparing with the diagnosis of 21 dermatologists, they found this deep neural network. The diagnostic accuracy is comparable to that of a human doctor, above 91%.

Deep learning adds medicine to medicine

In China, skin cancer is not a particularly attractive member of the cancer family because the incidence of skin cancer in yellow people is lower than in whites. But in the United States, skin cancer is one of the most common cancers. About 5.4 million Americans suffer from skin cancer each year. In the case of melanoma, if the treatment is detected and treated at an early stage within five years, the survival rate is around 97%; but in the advanced stage, the survival rate drops to 14%. Therefore, early screening is a matter of life and death for patients with skin cancer.

Under normal circumstances, after coming to the hospital or clinic, the doctor will conduct clinical screening based on visual diagnosis, and then perform dermascopic examination, biopsy and pathological diagnosis on the suspected lesions.

The doctor uses a dermatoscope to check.

But for a variety of reasons, many people do not go to the hospital in time for some small symptoms on the skin. Therefore, artificial intelligence-based home portable skin cancer diagnostic equipment will greatly improve the screening coverage of early skin cancer and save more lives. However, the diagnosis of cancer, the difference is a thousand miles, artificial intelligence can be qualified for the task of screening melanoma from ordinary sputum? The joint research team at Stanford University concluded that the diagnostic accuracy of machine-based doctors based on deep learning is staggering.

“We realized that it was feasible, the machine could not only do it, but it could be as good as humans,” said Sebastian Thrun, assistant professor of Stanford Artificial Intelligence Laboratory. “At this time our thoughts have completely changed. We said, '瞧Well, this is not just a student assignment, it may be beneficial to all human beings."

This visual processing algorithm is based on the deep learning of the moment, that is, training the machine to accomplish certain tasks through a large amount of data as an example. Recently, deep learning has not only shined in visual processing, but also has fruitful results in other different fields. For example, Google's Go AI Alpha Dog defeated World Go Champion Li Shishi after learning 30 million human chess. In the machine learning process, developers no longer need to code the problem solving method, but let the computer "learn" the solution by learning the sample data. In the case of skin cancer diagnosis, researchers no longer need to summarize some of the regular features of skin cancer in order to teach computers, but rather summarize the patterns themselves.

Silicone Endotracheal Tube

Features

Silicone Tracheal Tube,Reinforced Endotracheal Tube,Endotracheal Tube Cuffed,Reinforced Tracheal Tube

Hangzhou Trifanz Medical Device Co., Ltd , https://www.cfzmed.com