AI medical imaging is one of the fastest landing directions in the medical field

No one will question the significance of AI in promoting medical development.Among them, AI medical imaging is one of the fastest landing directions in the medical field.

The suzhou institute of medicine and industry of the Chinese academy of sciences, together with a research team from lishui central hospital and the second affiliated hospital of soochow university, has conducted a new study.The results showed that se-densenet, an artificial intelligence system used in conjunction with medical imaging, combined with enhanced mri images, could grade patients for cancer in a noninvasive manner.The team said it will use the technology in its developed navigation system for liver cancer ablation planning to assist in more accurate surgical planning.

Capital accumulation, policy support, relatively easy access to radiological imaging data, the domestic imaging doctor gap is huge...Many necessary and sufficient conditions are driving AI medical imaging forward.However, there are still many obstacles in the way of commercialization.

In medical diagnosis, the value of imaging is irreplaceable. 90% of medical data require doctors to judge pathological conditions, surgical plans and drug risks through imaging.However, in clinical application, image interpretation is highly dependent on doctors' experience and is highly subjective. Therefore, it is an important research direction to seek objective and effective evaluation methods.It is of great clinical significance to use "medical image +AI" to obtain more comprehensive lesion information and reduce the probability of missed detection.

The development of AI medical imaging is rooted in data.Data is the core data for deep learning algorithm.The uniqueness of Chinese healthcare forces ai companies to work deeply with hospitals.

In China's health system, hospitals are relatively independent with independent data. Hospitals in different regions follow different policies and management, and China's medical data control policies are not clear.This poses a number of difficulties for AI healthcare companies.In order to get valuable data from the hospital, let the doctor help machine learning in the research and development process, and let the hospital allow the product to enter the market for trial.In the actual product application process to get feedback.Technology companies have to spend a lot of money and manpower to fix hospitals and doctors.

In addition, there is a lack of standardized high-quality training sets at the present stage in China, which makes the data training set standards adopted by various artificial intelligence enterprises varied and the system deviation is relatively large.Each hospital has different medical procedures and policies, and companies generally need to customize their products separately, which increases the pressure of research and development and financial pressure on AI medical companies.

This is reflected in the direction of the company.

At present, most of the domestic companies engaged in AI medical imaging focus on pulmonary sarcoidosis, because of the large number and prevalence of cases and the relatively uniform response of patients, which is a relatively easy direction to overcome.But the most important reason is that there is a public database of pulmonary nodule image data. Any company that masters AI algorithm can run a model out of this database.But in other diseases, it is difficult for technology companies to get a large amount of data, and for AI companies, only a small number of hospital data does not make much sense.

The practicality of the product was questioned

In addition to the difficulties and enormous pressure in the process of promotion, some scientific research companies have been put off.In terms of product landing application, at present, the results achieved by AI are far from reaching the expected level, facing the challenge from professional doctors.

At present, AI medical imaging is basically a model based on image annotation of a single disease, and there are no products that meet the clinical use scenarios. The products are concentrated in a few diseases, and it is difficult to cover all medical imaging problems.The most important is the practicality of the product, that is, the accuracy of AI film reading.In clinical practice, many startups are only 50 percent accurate.It is difficult to gain the trust of hospitals and doctors due to the inaccuracy of visual identification and the inconsistency between self-reported product performance and actual test data.The issue is also common within the medical imaging industry including IBM Waston.

The market is booming, but it is too early to make money

Ai medical industry heat is rising, image as the recognized image recognition technology in the medical field the most direct application, set off a wave of entrepreneurship.According to incomplete statistics, more than 40 startups have entered the field of artificial intelligence medical imaging.In addition to the highly vertical artificial intelligence medical companies, Internet giants have also entered the market, affecting the pattern change in the field.

According to Global Market Insights, medical imaging and diagnostic technology will become the fastest growing industry in the smart medical field from 2017 to 2022. It is expected to reach 25 billion us dollars by 2024, with a growth rate of over 40%.

The reality is not so rosy.Even IBM Watson, the leader, has not yet reported profitability for its Watson for Oncology.

In February 2017, the ai healthcare bellwether program MD Anderson announced the end of its partnership with IBM Watson, which was seen by many as a setback for ai in healthcare.One reason is that the Oncology Expert Advisor (OEA), a clinical decision-making system supported by IBM Watson, the system they are collaborating on, has not been put into clinical use.

Compared with Watson, a number of domestic ai medical imaging companies are still in the application stage of disease screening, that is, to determine whether there is a certain type of disease in the image, and do not have in-depth analysis ability for specific symptoms of the disease.For example, lung image recognition may help doctors reduce some workload in clinical practice, but it is of little help to doctors and application scenarios, and its value is relatively low.As a result, hospitals and doctors are reluctant to pay for it.The cooperation with the hospital is mostly to provide product trial, receive no money.Without revenue streams and scenarios, business models are unhealthy.

But more painfully, monetization is something AI medical imaging companies have yet to think about, and the priority may be higher "surviving the intense competition."

Google, IBM, Intel and other international giants as well as domestic BAT and other technology giants have stepped up their layout to help start-ups get rid of the "big fish eat little fish" fate.Even if the giants don't consolidate the market aggressively, the competition among startups is fierce.

Summary:

Thanks to the deep learning technology of computer vision, there are many excellent AI startups in China, such as kuang shi technology, shang tang technology, video++ and cloud technology, which take the lead in landing AI on various runways.After years of development, AI medical imaging has become a key field of AI enabling applications.As the earliest and most competitive battlefield, how far is the commercialization of AI medical imaging?

The rise and bubble of the industry is the inevitable path of any new technology wave.Artificial intelligence enters bureau medical treatment, how to break through application close, still need to wait.