AccuRad: Boosting Diagnostic Efficiency with AI

AccuRad builds an AI-aided medical diagnosis system based on Intel® architecture.

At a Glance:

  • AccuRad has been focusing on core medical imaging technologies for twenty years.

  • AccuRad selected Intel® Xeon® Scalable processors and other cutting-edge products and technologies to create highly efficient and intelligent diagnostic capabilities to help medical institutions bolster the efficiency and quality of their diagnoses. The close cooperation resulted in AccuRad’s AI-aided medical diagnosis system receiving positive feedback from customers for its screening time, smart report writing, and other improvements.



The burgeoning development of medical imaging technologies in recent years has helped doctors examine their patients in greater detail and with better accuracy. While this is good news for both the doctors and the patients, these technologies have faced several serious challenges hindering them from unleashing their full potential.

One of the main challenges faced in China has been significant differences in the imaging systems technology and talent pool in medical institutions across different regions and tiers due to unequal resource allocation, resulting in different ability and rate of using imaging systems. Another challenge lies in the fact that many of the medical imaging devices are not yet interconnected or interoperable and are therefore unable to provide comprehensive diagnosis information to doctors due to inadequate data sharing. Even as these two challenges are resolved, it is still time consuming and energy intensive work for doctors and specialists to manually interpret films and analyze the huge amount of medical imaging data available. It is also often a challenge for doctors to maintain peak productivity and accuracy after long hours of work.

To tackle these challenges, Xi’an AccuRad Network & Technology Co., Ltd.* (“AccuRad”), a company that has been focusing on core medical imaging technologies for almost 20 years, is now looking to integrate its core medical imaging technologies and products with advanced technologies like cloud computing, big data and AI. The target of this initiative is to create highly efficient and intelligent diagnostic capability to help medical institutions bolster the efficiency and quality of their diagnoses.

AccuRad’s solution draws on multiple enhancements. First, AccuRad links massive medical imaging data from different equipment through its Yizhen Cloud to take advantage of the innovative medical equipment IoT technology AMOL*. Then, AccuRad delivers powerful big data processing of imagery through its @iMAGES core engine* for medical imaging processing and analysis. Finally, AccuRad builds Cloud IDT services* to bring AI technology to bear on the processing and analysis of medical imaging data.

For many years, Intel has been at the forefront of medical digitalization technology and AI innovations. In the process of creating and optimizing its solution, AccuRad teamed up with Intel to use Intel’s brand-new Intel® Xeon® Scalable processor and other cutting-edge products and technologies. With Intel’s support, AccuRad completed the migration of Cloud IDT services to an Intel® architecture-based platform and deployed and optimized TensorFlow* and other AI technology frameworks. The close cooperation between these two companies resulted in AccuRad’s AI-aided medical diagnosis system to achieve positive feedback from users for its screening time, smart report writing and other indicators.

Compared with traditional PACS systems, AccuRad’s new system focuses more on real-time demonstration of the data values contained in medical imaging using unique data cloud high-performance cluster computing. This new system also aims to improve diagnostic efficiency by introducing AI capabilities. Efficient implementation and operation of the cloud platform and AI solutions is dependent on powerful software and hardware platforms. This is where advanced Intel architecture-based products and technologies such as Intel Xeon Scalable processors give us a powerful performance boost.” —Yedong Huang, founder Xi'an AccuRad Network & Technology Co., Ltd.

Hyper-Connectivity Erases Discrepancies in Medical Imaging Analysis

Medical imaging technologies can significantly enhance the diagnosis process. However, while the right hardware is readily available, the same cannot be said for the software. It could be argued that to fully exploit the advantages of medical imaging technologies requires radiologists to possess not only professional knowledge in clinical medicine and medical imaging, but also expertise in radiology, CT, MRI, ultrasonics, etc. and at the same time they must be able to use all the various diagnostic imaging technologies to diagnose diseases.

In such circumstances, although medical imaging equipment may be widely employed by medical institutions, in some remote areas or even primary medical institutions there can exist a situation where, despite having the relevant equipment, there is no one with the expertise to read the output.

Figure 1 Yizhen Cloud Linking and Converging Medical Devices

In addition to this, since the information-based systems of the various medical institutions are independent of each other with no common standards for imaging data, the Picture Archiving and Communication Systems (PACS) of each institution can be considered an information silo. As the medical data they store are barely connected, some patients cannot get effective disease analysis in grass-roots hospitals in remote areas and are required to take repeat examinations after travelling long distances to hospitals with adequate facilities.

To resolve this problem, AccuRad introduced Yizhen Cloud which can link all related medical equipment as well as courses of medical services via the cloud. It does so with the help of the medical equipment IoT technology AMOL. As seen in Figure 1, a medical institution’s devices located in different departments, such as the Radiography Center, Pathology Center and Ultrasonography Center, can be converged via Yizhen Cloud. Based on this, AccuRad establishes the capabilities and applications including Accurate Full Medical Technology Operation & Collaboration System, Regional Medical Collaborative Platform, Clinical Imaging Research Platform, U. Doctor and Doctor Social to satisfy the medical imaging data processing needs of medical institutions of all levels via Software as a Service (SaaS) cloud.

Citing the example of Accurate Full Medical Technology Operation & Collaboration System, by accessing Yizhen Cloud, medical institutions at all levels can obtain cloud storage features for multiple devices and massive data, possess the cloud computing abilities for real-time processing and high-speed analysis and achieve a range of cross-terminal and cross-platform functional applications. With Yizhen Cloud, medical imaging specialists from large and medium-sized facilities may process imaging data transmitted from different regions anytime, anywhere and conduct collaborative consultations for complex diseases. Yizhen Cloud can therefore be seen to represent an effective way to achieve efficient sharing of medical resources.

“Intel® architecture-based Yizhen Cloud can truly help establish new, mutually beneficial relationships among patients, doctors and hospitals,” says Mr. Yedong Huang from AccuRad, “On the one hand, it provides doctors and patients a channel for remote diagnosis and treatment; on the other hand, it allows doctors to obtain the latest medical data and share their experience among themselves. At the same time, hospitals are able to conduct efficient communication for collaborative consultations, joint medical research and related activity.”

Yizhen Cloud enables the intercommunication and consolidation of medical imaging data, allowing medical institutions to avoid issues such as an excessive number of examinations and repetitive treatments. Yizhen Cloud also breaks down the walls of data silos to establish borderless medical connectivity and improve the quality of medical services. As data builds up and is analyzed, data utilization is further increased to support clinical decisions and launch new medical R&D. Meanwhile, processing data in the cloud also enhances quality control during the diagnosis and treatment process. Leveraging Yizhen Cloud, medical institutions can easily initiate quality control such as data quality control, content report quality control, diagnosis result quality control, clinical treatment quality control and the tracking of rehabilitation for chronic diseases.

A Powerful Engine That Empowers Medical Imaging Processing and Analysis

Migrating data to the cloud is the first step taken by AccuRad to build an AI-aided medical diagnosis system, however, the key issue is how to make use of the big data once in the cloud. Leveraging the powerful performance of Intel® architecture, AccuRad developed the @iMAGES core engine to allow cloud computing for medical imaging processing and analysis. This core engine enables high-speed real-time computing and processing of the imaging data stored on Yizhen Cloud.

As far as AccuRad is concerned, imaging is the result of computing rather than the result of “transmitting.” Thanks to higher parallel computing power delivered by the latest generation Intel Xeon Scalable processor, @iMAGES can rapidly perform multidimensional reconstruction of imaging data transmitted from remote sources.

To illustrate this, consider Positron Emission Tomography CT (PET-CT). In PET-CT, the PET scanning device provides detailed information including functions and metabolism of patients and the CT machine provides the precise anatomical position of patients. PET-CT is used for early detection and diagnosis of various diseases and since PET-CT examination is, in fact, a combination of two different methods, imaging fusion capability is crucial for the quality of examination results.

Figure 2 Cloud PET-CT Fusion

As seen in Figure 2, with the computing power delivered by Intel architecture-based platform, @iMAGES is able to offer remarkable cloud-based PET-CT fusion capability which not only provides thermal imaging based on morphology and functions, but also performs semi-quantitative Standardized Uptake Value (SUV) analysis on imaging for the subsequent identification and quantitative analysis of tumors and other illnesses.

@iMAGES is also able to offer powerful computing and processing of other types of medical imaging. For instance, in Functional Magnetic Resonance Imaging (fMRI), the @iMAGES engine can rapidly execute functions including diffusion, perfusion and nerve tract imaging. In angiocardiography, @iMAGES is capable of quantitative analysis of coronary artery segmentation, coronary axis, as well as coronary artery and plaques.

AI Delivers a Powerful “Brain” for Medical Diagnosis

To ensure that the AI-aided medical diagnosis system can truly improve medical institutions’ diagnosis and treatment ability, AccuRad and Intel cooperated closely. Working together to innovate has resulted in noteworthy progress in AI-aided medical diagnosis and treatment using the brand-new Cloud IDT services and the massive data collected and processed by Yizhen Cloud and @iMAGES.

Taking lung cancer as an example, during the early stages it is often presented as asymptomatic pulmonary nodules that can easily be overlooked. However, early diagnosis of pulmonary nodules (benign or malignant) can effectively lower the death rate for the disease. Because the miniscule pulmonary nodules tend to escape detection by the human eye, the disease is usually in the intermediate-terminal stage by the time it is diagnosed. In such circumstances, patients have lost the optimal window for effective treatment.

Today, aided by Cloud IDT services, the quantitative sensitivity (detection rate) of AI-aided diagnosis for a low-dose CT pulmonary nodule is as high as 95%. Screening time is also brought down to under six seconds from over 10 minutes previously when screening was performed manually.1 After identifying pulmonary nodules with AI, doctors will then perform further diagnosis for confirmation. This significantly increases the diagnostic efficiency and accuracy. The massive data held on Yizhen Cloud provides many training samples for the AI detection models to continuously upgrade the system’s detection ability.

Figure 3 AccuRad’s AI-based Medical Knowledge Graph

AccuRad’s AI-aided diagnosis system is planned to have hundreds of AI detection models for the various physiological systems of the human body in the future. Currently, its AI-based medical knowledge graph draft has already defined the association relationships of about 984 types of diseases and medical test data.1 At the same time, the system provides the “intelligent report assistant” function based on Natural Language Processing (NLP) to help doctors write high-quality examination reports more efficiently.

Intel’s Advanced Technologies Empower the Intelligent System

As AccuRad’s strategic partner, Intel offers wide-ranging technological support for the AI-aided diagnosis system. Apart from delivering the brand-new Intel Xeon Scalable processors, Intel has also helped AccuRad complete the migration of its Cloud IDT AI services to the Intel® platform and optimize the AI frameworks.

Intel Xeon Scalable processors, Intel’s newest generation of processors, feature powerful general-purpose computing power and provide the AI-aided medical diagnosis system with the parallel computing ability it needs. The massive imaging processing and AI processing involved in the system make a high demand on parallel computing ability. Intel® Advanced Vector Extensions 512 (Intel® AVX-512) integrated in Intel Xeon Scalable processors is the key technology that enhances the execution efficiency of Single Instruction Multiple Data (SIMD).

Intel Xeon Scalable processors can consolidate system application workloads because the processor excels at both general-purpose computing and parallel computing. According to tests, in terms of processing power, two servers with this processor can support up to 2.5 times the virtual machines compared with the original platform. This can significantly lower the Total Cost of Ownership (TCO) for users.2 “Initially our system used a GPU board for rendering and another GPU board for AI computing, while a general-purpose processor was used for business processing. The cost was high, and maintenance was complex. Now, we only need to deploy Intel Xeon Scalable processors and everything is resolved,” said Mr. Yedong Huang, “We are now planning, therefore, to migrate all tasks deployed on different hardware platforms to the Intel Architecture-based platform.”

Based on superior hardware performance, Intel’s optimizations for AI frameworks such as Caffe* and TensorFlow* further augment the capability of AccuRad’s AI-aided diagnosis system. The RFCN model optimized for Intel® technologies saw model optimization cropping and merging performance improve by nearly 30%. Further performance enhancement of 40%-50%2 is possible by optimizing the OpenMP* multithreading implementation solution.

Additionally, the introduction of Intel® Math Kernel Library for Deep Neural Networks (Intel® MKL-DNN) provides the system with another step towards overall intelligence. This tool enhances the performance of AI models in the following three ways:

  • Using Cache Blocking* technology to optimize data caching and increase data hit rate.
  • Conducting parallelization and vectorization optimization on common operators in the neural network.
  • Adopting Winograd* algorithm-level optimization.

The coordination and fine-tuning of these Intel® software and hardware elements enable AccuRad’s AI-aided medical diagnosis system to deliver a superior performance and win praise from its users. The test data from first-line deployment indicates that, in the case of executing RFCN model using data from a single chest DICOM image, the Intel® Xeon® Gold 6148 Processor reduces processing time by 10%2 compared with a mainstream GPU.

Figure 4 Processing Delay Comparison of Executing RFCN Model Using Data from Single Chest DICOM Image


Based on Intel’s cutting-edge AI infrastructure products and technologies, integrated with AccuRad’s innovative cloud computing, medical imaging processing/analysis and AI technology, the AI-aided medical diagnosis system is now live. To date, it has been connected to over 1,000 medical facilities, seeing breakthroughs in areas including healthcare collaboration, real-time computing of medical imaging data and AI analysis of medical visual data, and garnering much positive feedback in its deployment and implementation.

Building on this success, AccuRad and Intel will continue to deepen their cooperation with plans to amalgamate more advanced Intel products and technologies with medical informatization. AccuRad and Intel will also make joint efforts to deploy AI solutions in areas like collaborative medical research, medical big data analysis, medical health management, disease surveillance and imaging analysis etc. to further advance precision medicine and intelligent healthcare.

Download the PDF ›


1Data cited from AccuRad’s unpublished research program and testing results.
2Data cited from AccuRad’s internal performance test results supported by Intel.