Race with computing power and time, medical artificial intelligence protects the light of life

Time is life, and no moment can better confirm this statement than the rescue process. Every minute counts, taking the first step

Time is life, and no moment can better confirm this statement than the rescue process. Every minute counts, taking the first step... shortening the time and improving the efficiency of diagnosis and treatment during the rescue process may change the fate of some people.

45 year old Wang Dacheng is a community worker. One day, he developed symptoms of impaired left hand movement that gradually worsened, but he didn't pay attention and still held onto his position. During the night shift, he suddenly became paralyzed and was discovered by his colleagues who took over the shift. He was urgently sent to the First Hospital of Jilin University and entered the green channel for stroke treatment. After preliminary physical examination, the doctor determined that Wang Dacheng had a relatively long history of illness and needed to be treated as soon as possible.

In this process, artificial intelligence is involved, assisting doctors in completing image data analysis within 3 minutes, and integrating multiple disciplines to come up with a treatment plan for mechanical thrombectomy surgery on patients. With maximum effort, patients are pulled back from the brink of lifelong paralysis and death. With the support of green channel and platform technology, they have won this "battle for life".

3 minutes! Race against time, artificial intelligence participates in "life borrowing"

The treatment of stroke is a race against time. For every minute of delay in the treatment of stroke patients, 1.9 million brain cells are damaged. Therefore, the treatment of stroke has a very strict time window. It is not easy for a considerable number of doctors to make a treatment plan within the window period. Especially after 6 hours of onset, most areas of the patient's brain tissue are damaged, and the diagnostic ability of grassroots doctors is insufficient. In addition, there is a lack of reliable and accurate automated evaluation tools in clinical practice, which makes it difficult to accurately identify the patient's core infarction area and determine the brain tissue areas that can still be saved. Many stroke cases may have a diagnosis time of up to 100 minutes, resulting in delayed treatment rates that cannot be improved. Since 2015, the reperfusion treatment rate for acute cerebral infarction in China has been much lower than that of European and American countries. The actual implementation rate of intravenous thrombolysis for acute cerebral infarction patients with no contraindications within 4.5 hours of onset is only 22.9%.

The application of artificial intelligence medical imaging assistance systems developed by Inspur and Metabrain Left Hand Partners in stroke may change this situation. Usually, the time for diagnosing stroke to the imaging department is only 15-20 minutes. This system can accurately detect the infarct lesion and automatically, intelligently, and quickly segment the blood supply area, watershed area, and structural area according to different medical needs, thus achieving rapid qualitative and quantitative analysis of the infarct lesion. The accuracy of the multi region synchronous segmentation model reaches 97.5%, and can provide reference diagnostic reports within 3 minutes..

Advanced medical treatment sinks into artificial intelligence, benefiting more people

China is a major stroke country with over 3 million new stroke patients each year

[1] . Among the patients with nervous system disease admitted to hospitals in China, stroke accounts for 66.5%

[2] . Patients face health threats, and doctors also face challenges.

Diagnosis and treatment of stroke are both technical and experiential. It often takes 10 years to cultivate a highly qualified stroke doctor. China has a vast territory and significant regional disparities in medical development. Such doctors are often concentrated in tertiary hospitals in developed regions, and the level and experience of doctors in the vast peripheral areas urgently need to be improved. This is also the direct reason for the unsatisfactory treatment of stroke in China.

The emergence of artificial intelligence medical imaging assistance systems is improving this situation. The detailed diagnosis of stroke mainly relies on imaging. Artificial intelligence medical imaging assistance systems can fully solidify the best doctor's diagnostic experience into algorithms and solutions, and solve the stroke diagnosis problem in grassroots medical institutions through the "equipment going to the countryside" approach. Compared to doctors going to the countryside, the feasibility and efficiency are much higher.

Modern technology has opened up new avenues for achieving medical equalization.

AI healthcare, computing power, algorithms, and data are indispensable

The AI aided analysis medical products used in Wang Dacheng's case are trained based on the clinical image data of tens of thousands of cases in the world's leading hospitals for neurology research and treatment. The hospital not only has a massive amount of clinical imaging data, but also has a very high level of treatment. The recurrence rate of stroke reperfusion treatment in China is 25%, while the international level is 7%. The recurrence rate of treatment in this hospital is far lower than the international level.

The artificial intelligence model deeply learns the diagnosis and treatment techniques of senior clinical experts in hospitals, supports multimodal image automatic analysis, including CTA (CT vascular imaging)/CTP (CT perfusion imaging) analysis, magnetic resonance analysis, etc., meets the equipment requirements of different levels of hospitals, and can assist clinical medical students in making better treatment decisions.

The two dimensions of data and algorithms have already exerted their power, but without strong "momentum" to drive their operation, they cannot achieve efficiency. Inspur's intelligent computing power provides fundamental guarantee for AI to truly achieve medical assistance.

Computing bottleneck, AI participation in diagnosis truly enters clinical practice

What most people may not be aware of is that artificial intelligence algorithms are not only written by programmers, but also consume a lot of computing power to "calculate" them. After the development of artificial intelligence models is completed, a large amount of data training is required. The model developed is equivalent to a baby, and the training process is the process by which the baby grows up to become an expert.

The medical imaging system requires hundreds of GB of data for each training session, and the training platform provided by Inspur Information can complete nearly billions of training sessions per second. It should be noted that people need to constantly learn and grow, and the AI system also needs to learn the latest cases and keep learning. This also means that the AI system may need to "lifelong learning" and continue to require huge computing power to "feed".

Of course, in practical application deployment, this system also requires a powerful computing power platform to reduce computing time. Although deployment situations vary, the peak computing power level of each system is also at a super high level of billions of times.

Inspur Information not only solves computational power problems, but also challenges in the development and deployment of solutions. The development and training of AI systems require collaboration among multiple people. So, Inspur not only provides an AI computing power platform, but also an AI resource management platform, AIStation, for unified and efficient management of computing power resources. It supports dozens of partners to use the computing platform at the same time, significantly improving resource utilization and training efficiency. The GPU utilization rate has increased from 30% to 75%, greatly saving computing costs and improving efficiency: the main model training speed has been increased by more than 10 times, The training time has been reduced from over 2 weeks to 2 days.

With high-quality data sources, leading algorithm models, and strong computing power platforms, and through continuous analysis and training of new cases, this artificial intelligence medical imaging assistance system is constantly self iterating and upgrading with the support of computing power, improving accuracy.

The impact of AI on modern medicine goes far beyond this. Inspur Information has also collaborated with Northwestern University in the United States to develop an artificial intelligence NLP system to identify radiological examination reports that require follow-up. This achievement was published in the sub issue of the New England Journal of Medical Catalysts (NEJMCatalyst).

The purpose of medical treatment is not only to save patients, but also to restore their health, return to society, and return to life.

The shortening of time is an important manifestation of the effectiveness of technology, and it is still this that brings Wang Dacheng's story to a successful conclusion. At the same time, China's medical resources are scarce and unevenly distributed, and the grassroots medical force is weak. The cross application of clinical and artificial intelligence transforms high-quality clinical data into universal experience, assisting grassroots doctors in quickly and accurately completing image data analysis, reducing missed diagnosis and misdiagnosis, and improving doctor work efficiency to benefit more patients... In the process of promoting medical homogenization in China, it benefits the lives and health of the people on a larger scale.

Information source:

[1] Chinese Guidelines for Clinical Management of Cerebrovascular Diseases, page 289

[2] The research and application of artificial intelligence in stroke diagnosis and treatment: dawn is emerging, and there is a long way to go; Chinese Journal of Stroke, March 2020