How far is the intelligent imaging diagnosis to come true?
更新时间：2017-08-02 17:04:02•点击：825829 • Industry Views
Peoples always heard that “artificial intelligence replace doctor” or “the AI accurate rate more than doctor”. Meanwhile, people feels “The ideal is beautiful, the reality is very cruel” and we own the ambitious words and hopes, but how about the reality?
It is said that data was dominated in 21 century. Some people compared arithmetic to engine and data to petroleum, other people emphasized that indispensable three elements to build intelligent image is industry data, expert resources, and core technology. The importance of data cannot be overemphasized. Furthermore, We will take the image data as the tool, and the operation of the smart imaging company as our aim to study and learn the real daily life of the smart imaging company.
Data terminal: Insure quality and quantity, the more the better.
Despite third party image center exists in China, but most of the medical image data source of hospital. It is estimated that the large 3A hospitals will produce more than 10t image data in a year. WuBo said, YIYUAN intelligent CEO, "the single hospital image data stock is large and increase more than hundreds of new cases is common every day."
In HIS, PACS system is responsible for medical image collect, data update and storage, also can analyse and deal with image. Meanwhile, the different PACS system will connect with each other by DICOM international standard mode.
Overall, the hospital image data numerous ,large and standardization which can easily used by machine. Many insider believed that the intelligence health care will first to achieve commercialize.
Yang Xiujun, Shanghai Children's Hospital imaging department director, have said:“ Many medical imaging areas are particularly suitable for artificial intelligence or image recognition technology, and many companies at home and abroad to engage in this also to made some achievement.”
For AI system, the data that the more the better good need precondition. Under ensuring the quality of data, increase the number of is more meaningful. And judge the quality of image data, mainly depends on the AI company to build intelligent diagnostic products for clinical purposes. In addition, for intelligent imaging diagnosis, the image data needs to be associated with more accurate diagnosis and late outcome association, otherwise the useless data comes in and then out of.
Southern Medical University associate professor Liu Zaiyi has said “We have many data and our image department generate numerous data everyday, but how many data are available? Less than 1 percent and there has many error information”. He added, “ there are no way to manage the data standardization and clinical information frequently made mistakes.
Data acquisition: “win-win” cooperation
The image data own hospital, the intelligent image company how to get it?
Medical data is a resource, meaning it is valuable. If we want to get valuable things and the simplest logic is to buy it, which is rich and powerful IBM's strategy. In 2014, after acquiring Phytel and Explorys, a large data provider for health care analysis, in 2015, Merge Healthcare, a medical imaging and clinical systems provider, was acquired for $ 1 billion, with 8500 customers, including the US federal government, state government agencies, employers, health insurance, hospitals, etc., and 300 million patient data.
In China, the 3A hospitals have the majority of image data, but the image data can not leak out. Therefore, AI companies and hospitals to seek "cooperation" has become a possible path. Generally speaking, AI companies will choose to cooperate and developed with the hospital. According to this not only attains desensitization data and industry experts but also achieves harvest of the product grinding scene. As for the cooperation model, they have their own characteristics.
Data processing: “Only artificial, no intelligence”
Just like the machine learn AI modeling, medical image data processing should also experienced data labeling, cleaning, cutting, and then modeling and tuning and so on.
In dealing with the technical problems of video data, according to WuBo introductions, medical image data portrayed the body organs, which different with the naked eye to identify the birds and other common images. The imaging principles and visual characteristics are also different with that. In-depth study model need to deepen and transform.
But the medical image data processing is special in that annotate the data will takes more time and the threshold more higher. In the doctor's diagnosis, the image is only a reference information, and ultimately depend on the pathological diagnosis and other information to diagnosis and confirm. So numerous data collection is valid and useful data for built a smart diagnostic system. AI companies need to open up different systems as much as possible, then to integrate all the patients information and relevant information. However, it is actually more difficult. (Source: LeiFeng network)
