AI is a medical image that can be said to be a savior. The combination of powerful computing power and learning ability allows AI to quickly find patterns from countless medical imaging films, transforming doctors' decades of reading experience into technical tools that can be copied and popularized. In addition to greatly liberating the productivity of doctors and making the nervous medical resources slightly relieved, they have begun to surpass humans in the discovery of certain diseases. As in recent reports, Google and a medical institution in the United States cooperate to use artificial intelligence to diagnose and monitor. Breast cancer, in the monitoring of metastatic breast cancer, Google artificial intelligence system has achieved 99% accuracy, has exceeded the level of human pathologists.
Only the recognition on the image can bring such a huge change. From this year on, the improvement of computing resources has made AI more and more enter the video field. What is the medical analysis? influences?
Using video to analyze Parkinson's disease? May not be as magical as you think
In fact, in the AI ​​analysis video for intelligent assisted diagnosis and treatment, relevant application cases have emerged. A few months ago, Tencent Medical Artificial Intelligence Lab launched a new AI-assisted diagnostic technique for Parkinson's disease called the Parkinson's Disease Motor Function Assessment System. The patient completes the action according to the Parkinson criteria and takes a video. The evaluation system can help the doctor to judge whether the patient has Parkinson's disease in 3 minutes through the completion of the action, and the doctor needs a lot of communication and 30 when he is alone. Minutes around the eye.
Users do not need to wear complex sensors, just use the ordinary mobile phone camera to complete the shooting. Remote diagnosis can be done in homes, nursing homes, etc., greatly improving the convenience and efficiency of Parkinson's diagnosis.
Of course, the AI ​​analysis video is still only a case for Parkinson's assisted diagnosis. The reason is that there is a strict set of criteria for the diagnosis of Parkinson's internationally, called UPDRS (Unified Parkinson's Disease Rating Scale). There are strict criteria for the action part, such as the flexibility when standing up, the degree of tremor when the hand moves at rest, and so on.
In other words, Parkinson is a disease that can be "structured." There are very clear criteria for how to visually judge disease. This kind of formatable disease is very rare. We can't judge colds and fevers through human movements, and we can't judge the sprained bones by action.
Therefore, the scope of application of intelligent auxiliary diagnosis through motion analysis is still slightly narrow.
A sense of video intelligence analysis that may make emergency medical care save more lives
However, discussing the application value of video intelligence analysis to medical care cannot be discussed separately for a certain disease. In terms of efficiency and cost, medical video intelligence analysis is harder than simple image analysis. After all, most observable diseases look at static characterization rather than dynamics. For example, the wound of the skin is red and swollen. It is obvious that by taking a photo or even a few sentences, it is necessary to record a video.
But the medium of video has a feature that the image does not have, that is, the existence of "non-inductive". Video surveillance exists everywhere in our lives, but we are used to living in this kind of record. In this way, video intelligence analysis and treatment can turn medical care from passive to active.
Active medical treatment means that when we feel unwell, we will seek medical advice, go to the hospital, and use mobile medical apps. But many times, we have not chosen to take the initiative to heal, or even have time to heal. In particular, some sudden illnesses, if not found by others, are likely to cause irreparable damage to the patient without knowing it.
But combined with the presence of the camera and intelligent analysis of the video, it is likely to change this situation. For example, in a scene such as a nursing home, video analysis can accurately identify and alert if the old person is fainting in the absence of a surrounding person. Even through the analysis of the whole process of fainting, the basic disease direction can be judged, such as brain damage and body fracture caused by falling, or heart disease may be caught in the chest, so that treatment can be better, because there is no patient. Fainted and unable to understand the condition of the disease.
Today, with the increasing popularity of empty-nest families, this system of connecting emergency medical care through intelligent analysis of video brings not only economic benefits, but also the protection of human life.
Video emergency medical treatment that is difficult to implement, waiting for you on the road ahead of AI
The implementation of this emergency medical system is also facing a lot of problems.
The first is to identify the training of the model. Unlike medical images, which have detailed data models, emergency situations that require first aid are inherently vague. We often see news that some people feel unwell in public, sit down and take a break to leave. In many cases, humans can't see the difference. Will AI be weak? For more important questions, accurate model training often relies on a large amount of data training. The situation of emergency medical treatment itself is difficult to leave data - otherwise it will not cause so many tragedies.
The second is the boundary of privacy. The premise of video intelligence analysis is likely to be the ubiquitous camera. We are able to accept cameras in public because we trust the monitoring system of the entire country. But which side of the camera should be implemented in private, how users trust suppliers, and how data is monitored is a difficult problem to solve.
Finally, there are limits on computing power. Unlike general AI applications, video-based emergency medical care itself is "urgent." Many AI applications can wait until the data is uploaded to the cloud and return the results. Emergency medical treatment can not afford the risk: some people playing the glory of the king caused the network speed to slow down, and the power system suddenly tripped, resulting in no network. These sudden situations are not only life-threatening, but also bring intractable problems. ——The roommate downloading the movie takes up the speed of the network and causes me to be confused by the video emergency system when I am fainting. Finally, I am sick because of illness, and the roommate needs to be responsible for my death?
In fact, in this way, video intelligence analysis for emergency medical applications, the problem to be faced is the same as the current development of AI technology: the development of small data and even no data learning, training more accurate models with as little data as possible; Local computing, try to improve local computing power and converge the neural network model, let the analysis of data be done locally to improve the calculation speed and protect the security of the data.
This also illustrates a problem. AI should not only grow in the soil that adapts to itself, but also use a large amount of interest data for analysis to encourage merchants to sell more items or add a click to an article. Among the saline-alkali land that requires technological breakthroughs, it may be able to capture more value from the business.
Breast Pump Accessory,Bell Horn Hood,Nipple Protective Cover,Silicone Nipple Protective Cover
NINGBO YOUHE MOTHER&BABY PRODUCTS CO.,LTD , https://www.oembreastpump.com