GE Healthcare Launches End-to-End 'MR+AI' Platform to Tackle Key Challenges in MRI: Artifacts and Imaging Efficiency

According to Leifeng.com, on March 28, GE Healthcare unveiled its latest digital health innovation—the “Zhijian AI” platform, an end-to-end artificial intelligence magnetic resonance imaging (MRI) technology solution. Reportedly, the platform underwent a three-year research and development cycle and was trained on 100,000 raw imaging cases. It leverages AI technology at the core front-end stage of image acquisition to centrally address artifacts and improve imaging efficiency in MRI scans.

Zhao Xia, General Manager of the MRI Product Division at GE Healthcare China, stated that the Intelligent Simplified AI Platform will be deployed on the 1.5T SIGNA™ Creator MRI system. “Soon, our entire portfolio of MRI products will also incorporate this technology, empowering our full product line with AI and laying the foundation for advancing the MRI AI ecosystem.”

Compared to imaging modalities such as CT and X-ray, the development and application of AI technology in the field of magnetic resonance imaging (MRI) have been relatively lagging due to the complexity of imaging principles, acquisition processes, and image processing.

Professor Tian Jie of the Chinese Academy of Sciences, Chief Scientist of the National Key Basic Research Development Program (973 Program), stated that X-rays provide two-dimensional information, CT provides three-dimensional information, and magnetic resonance imaging can display functional information, resulting in a substantial increase in the volume of information.

Lu Guangming, Director of the Department of Medical Imaging at the General Hospital of the Eastern Theater Command and Vice Chairman of the Radiology Branch of the Chinese Medical Association, also stated that magnetic resonance imaging (MRI) provides multidimensional data across various sequences. The same anatomical structure or pathological lesion may exhibit different signal characteristics on different sequences. How to integrate these multidimensional information for lesion analysis is a challenge that all professionals in the medical imaging field need to address.

Liu Shiyuan, Director of the Department of Diagnostic and Interventional Radiology and Nuclear Medicine at Shanghai Changzheng Hospital and President-Elect of the Chinese Society of Radiology under the Chinese Medical Association, believes thatMagnetic Resonance Imaging (MRI) is the most complex technology in the field of medical imaging, representing the pinnacle of the pyramid. Due to the complexity of its imaging dimensions, it is challenging to develop deep learning models, resulting in limited breakthroughs in the field of AI.

Certainly, with the continuous application of artificial intelligence technology in the medical imaging industry, "high-end" MR equipment has also achieved significant improvements.

Tian Jie added that in the past, AI technology development in the field of magnetic resonance imaging has focused on two aspects:

First is the intelligentization of the MRI workflow (the MRI AI 1.0 era), such as automatic anatomical region identification and automated continuous scanning;

Second, various intelligent analyses and computer-aided diagnosis in the post-processing stage of image reconstruction (the AI 2.0 era of MRI), such as multimodal fusion of structural and functional imaging, can help improve image processing efficiency and provide more detailed imaging diagnostic information.

As mentioned above, a major challenge facing magnetic resonance technology lies in the “source” of the equipment:

In magnetic resonance imaging (MRI) examinations, 70% of the time is occupied by the imaging process, which can last from a few minutes to over half an hour. The prolonged diagnostic time has led to MRI appointment wait times at some large hospitals extending to more than a month.

Moreover, magnetic resonance images can produce artifacts.

Anyone who has undergone an MRI scan knows that doctors typically instruct patients to hold their breath and remain still before the examination begins. This is because natural peristalsis of internal organs or involuntary movements—such as intestinal peristalsis during abdominal scans or spontaneous swallowing during cervical spine scans—can create artifacts on the images, thereby affecting diagnostic accuracy.

In more severe cases, patients are required to undergo repeat scans or reschedule their examinations. Clinical practice data indicate that approximately 20% of magnetic resonance imaging (MRI) examinations fail due to motion artifacts.

“Issues such as ‘imaging speed’ and ‘artifacts’ have long plagued medical device manufacturers and clinicians. Hence, this is the rationale behind GE HealthCare’s launch of its Intelligent Simplified AI platform.”

By leveraging deep neural network algorithms and learning from over 100,000 cases of raw magnetic resonance imaging (MRI) data, GE Healthcare extracts features from the raw data acquired by each coil element during the imaging process, identifies and processes noise and artifacts, and eliminates impurities such as noise from the data, thereby providing high-quality raw image data.

Meanwhile, during the image reconstruction optimization process, AI-based algorithms are also employed to suppress image artifacts and significantly enhance the signal-to-noise ratio, thereby substantially improving imaging speed.

In terms of imaging speed, test results from GE Healthcare’s global magnetic resonance research and development team show that:

In shoulder joint imaging, the full-process AI Magnetic Resonance Creator uses the same field of view and parameter settings,Reduced imaging time from approximately 3 minutes to about 1.5 minutes, doubling the imaging speed；

In high-resolution intracranial time-of-flight (TOF) angiography, the same scanning sequence and image parameters were used.Traditional imaging scans take approximately 4 minutes and 50 seconds, while the Full-Process AI MRI Creator achieves an imaging speed of about 2 minutes and 20 seconds.Achieved higher-resolution and higher signal-to-noise ratio images in half the time.

In terms of artifact management, ZhiJian AI technology can effectively identify artifact signals, suppress the generation of various types of artifacts, improve scanning success rates, and reduce economic losses caused by repeated scans.

According to data from the China Association of Medical Equipment, in recent years, although the number of MR units per million population in China has increased significantly, rising from 3.3 units in 2013 to 6.2 units in 2017, nearly doubling, the overall market penetration rate of MR in China remains at a relatively low level, with considerable room for improvement compared to developed countries such as those in Europe and the United States.

Yu Chao, Managing Partner of Deloitte’s Life Sciences and Healthcare Consulting Practice, told Leiphone that the domestic MR market is divided into existing and incremental segments. From the perspective of the existing market, MR equipment has a lifecycle; next-generation MR systems offer faster imaging speeds and superior image quality, delivering significant improvements in operational performance.

From the perspective of the incremental market, the base-level market is gradually expanding, with an increasing number of primary-care hospitals upgrading their equipment tiers. In addition, the emergence of new business models, such as third-party imaging centers, is continuously expanding the incremental market.

Thus, it is evident that GE Healthcare’s launch of this new product aims to fulfill what Zhao Xia described as “the core responsibility of medical equipment manufacturers”—Leverage AI and digital technologies to expand the existing “installed base” market and penetrate the broader “incremental” market.

The ZhiJian AI Launch Event also hosted an online summit forum on artificial intelligence in medical magnetic resonance imaging.

Regarding the future development of artificial intelligence, Tian Jie believes that it mainly boils down to two factors: application-driven and technology-driven. Application-driven refers to identifying problems from clinical practice, utilizing AI technology to explore multi-dimensional information, and achieving more precise analysis in preoperative prediction and postoperative recovery; technology-driven, on the other hand, involves using AI to obtain more features, enabling doctors to see more clearly.

“High-quality data and artificial intelligence are mutually reinforcing, with demand and technology driving each other forward, thereby enabling vendors to better deliver clinical solutions.”

Notably, a significant value of AI-enabled devices lies in enhancing data quality and standardizing imaging capabilities, which is critical for big data infrastructure and the application of AI technologies.

As an expert in radiology,Lu Guangming to Leifeng.com(WeChat Official Account: Leiphone)It stated that, building on existing AI applications, manufacturers should next focus on enhancing the generalizability of AI technologies across different medical devices and ensuring the consistency of imaging data.

Director Lu believes that building a large-scale, high-quality database plays a significant role in disease identification, diagnosis, and assessment. The establishment of such a database is an ongoing process. “AI has provided us with excellent image quality; however, whether legacy images are compatible with current images based on 3.0T systems remains to be further observed.”

“Particularly for magnetic resonance imaging, factors such as varying scan parameters and different sequences can all affect data continuity. ‘We encounter many challenges, which require policy support and vendor assistance to jointly help us address these issues.’”

Li Jingli, Director of the Institute for Medical Device Control at the National Institutes for Food and Drug Control, stated thatChina’s first batch of artificial intelligence industry standards has seen two established this year, primarily consisting of foundational standards for general dataset requirements and terminology. Within the framework of these two foundational standards, research and development of standardized products will be carried out.

She also revealed that the National Institutes for Food and Drug Control (NIFDC) will solicit public comments in the second half of this year.

Original article by Leiphone; unauthorized reproduction is prohibited. For details, seeReprint Guidelines。