AI's involvement in diagnosis enters the clinic, and Ander's medical wisdom and wave work together

5;"> In 2020, the State Food and Drug Administration approved the issuance of the first medical AI software NMPA Class III medical device certification named "Image-Assisted Diagnosis", which means that the clinical medical field has entered a new era - more than just diagnosis for patients Only doctors, AI can also participate in image analysis, assisting doctors to make more accurate judgments.
The AI ​​software that got this three-category card is BioMind's "Tianyizhi" intracranial tumor magnetic resonance imaging-assisted diagnosis software under Ande Medical Intelligence.
Participating in the development of this software are Inspur AI Server AGX-5 and Inspur AIStation.

AI provides new ideas for improving disease diagnosis capabilities

The global cancer survival trend monitoring report released by The Lancet in 2018 shows that the overall 5-year survival rate of cancer in my country is 36.
0%, which is much lower than the 64.
0% in the United States.
The reason for such a large gap is largely due to the large room for improvement in early cancer screening in my country: on the one hand, Chinese residents generally lack the awareness of early disease screening, on the other hand, a large number of medical institutions in my country (especially Basic-level medical institutions) are limited by medical equipment, lack of senior medical personnel, etc.
, and it is difficult to detect and judge diseases in a timely and accurate manner from medical images.

To solve this problem, of course, we can rely on the continuous training and delivery of senior medical professionals to improve, but the professional medical personnel resources need to be accumulated for a long time and cannot be achieved overnight.
In this context, AI has demonstrated unique advantages.
It can help doctors quickly screen out medical images through features extraction networks, classification and segmentation networks, improve the accuracy of image analysis, shorten diagnosis results reporting time, and improve The diagnostic capabilities of the medical system.

Beijing Tiantan Hospital affiliated to Capital Medical University (hereinafter referred to as: Tiantan Hospital) and Beijing Ande Medical Intelligence Technology Co.
, Ltd.
(hereinafter referred to as: Ande Medical Intelligence) have jointly established the world’s first "Neurological Disease Artificial Intelligence Research Center".
Through the combination of medicine and industry, we will jointly launch the world's leading research, translation, and clinical exploration of artificial intelligence applications for neurological diseases.
The BioMind "Tianyizhi" MRI-aided diagnosis software for intracranial tumors, which obtained the first three types of NMPA certificates mentioned above, was born here.

Tiantan Hospital can use AI technology to sink its deep accumulation and leading advantages in neuroscience to primary hospitals.
The primary hospital is equivalent to introducing a senior expert.
Residents can get high-quality and personalized treatment plans at their doorsteps without squeezing into big cities or big hospitals.
This will greatly ease the difficulty in seeing a doctor.
Expensive question.

AI R&D computing platform is upgraded, Ande Medical Intelligence and Inspur work together to make efforts

During the development and training of AI models based on tens of thousands of clinical imaging data by Tiantan Hospital and Ande Medical Intelligence, they encountered three problems related to AI computing power.

First, the amount of CT/MRI image data is large, and a single case can reach the GB level.
The data scale of tens of thousands of cases puts huge pressure on the original computing platform, and the data reading and processing speed is very slow.
Second, in order to achieve accurate diagnosis of multiple diseases, the corresponding AI model is highly complex and computationally intensive.
The original computing platform is used for distributed training, which takes up to two weeks in a single time.
The third is that engineers need to develop and train corresponding models for various parts of the body.
There are many training tasks and many users.
One card is often required to handle multiple tasks.
The distribution of computing power resources is uneven, the utilization rate of computing power is not high, and development The environment deployment is complicated, and engineers maintain a set of development environment, which is not conducive to teamwork.

Ande Medical Intelligence hopes to upgrade the computing platform, while resolving the AI ​​computing power bottleneck, and at the same time, the unified and efficient management of computing power resources.
In response to this demand, Inspur provides an overall solution including Inspur AI server AGX-5 and AI resource platform AIStation.

The image reconstruction and deep learning models used by Ande Medical Intelligence engineers have large data volume and high complexity, and require high video memory and performance of the computing platform.
Compared with the original computing platform, AGX-5 has doubled the video memory, greatly shortening the throughput time of massive image data, and can support larger and more complex model training.
In addition, the AGX-5 stand-alone computing performance is as high as 2 PetaFLOPS, which is 3 times that of the original computing platform.
Through the high-speed interconnection of 16 AI chips, the parallel training speed of the model is greatly accelerated, and the training speed of the main model of Ande Medical Intelligence is increased by 10 Times more.

Inspur AIStation has assisted Ande Medical Intelligence in achieving resource management goals such as further improvement of computing platform resource utilization, simpler environment deployment, and more convenient teamwork.
On the one hand, engineers can apply for resources as needed and submit multiple tasks at the same time through the AIStation visual interface.
When resources are insufficient, tasks will be queued up, and resources will be automatically released after tasks are completed, so that they can make full use of idle time for training and improve resource utilization.
On the other hand, AIStation provides a variety of mainstream deep learning open source images.
Engineers can create development environments in seconds through the form of containers.
At the same time, the development environments created by the team members are visible to each other and can enter development and debugging to facilitate synchronization of work progress.
AIStation supports the simultaneous use of the computing platform by nearly 80 engineers of Ande Medical Intelligence, which significantly improves the resource utilization rate and training efficiency.
The GPU utilization rate has increased from 30% to 75%, and the training time has been reduced from more than 2 weeks to 2 days.

The close cooperation with the AI ​​computing platform has made progress in research and development and improved the accuracy of disease diagnosis.

In addition, Ande Medical Intelligence and Inspur also reached an ecological strategic cooperation agreement on Yuannao.
They will continue to integrate the superior resources of both parties, develop software and hardware integrated medical + AI solutions, and accelerate the implementation of applications to satisfy more medical institutions and more.
The demand for auxiliary diagnosis of multiple diseases promotes the sinking of high-quality medical resources, meets the people's growing demand for health services, and contributes to "Healthy China".

Source: Inspur Information