Jiufeng Medical: Empowering Grassroots Healthcare in Yuanzhou District with AI-Driven Remote Imaging Services

Low levels of informatization, coupled with shortages of funding and professional radiologists, are significant factors hindering the development of primary healthcare institutions. Telemedicine, interoperability, and artificial intelligence can help alleviate these challenges to some extent. Guided by national policies on tiered diagnosis and treatment and targeted poverty alleviation, Yuanzhou District—a revolutionary base area located on the Jiangxi-Hunan border—has collaborated with a technology company to develop its own unique model.

Yuanzhou District in Yichun City, Jiangxi Province comprises 22 townships and 10 sub-district offices, with a total population of nearly 1.1 million, of which approximately 70% are registered rural residents. Constrained by factors such as lagging transportation infrastructure, a weak economic foundation, and an inefficient urban administrative system, Yuanzhou District has long suffered from scarce medical resources and a widespread shortage of professional technical personnel, with the scarcity of imaging diagnosis specialists being particularly pronounced.

According to VCBeat, there are 24 township health centers in Yuanzhou District, but only three physicians hold valid qualifications for imaging diagnosis, including one re-employed retired doctor. This falls far short of meeting routine clinical needs. As a result, diagnostic reports for patients are often written by radiologic technologists or attending clinicians instead.

Non-standardized imaging reports directly lead to frequent misdiagnoses and missed diagnoses, and have also contributed to a certain degree of public distrust in local medical institutions. Many people prefer to travel long distances to seek care at large medical centers such as Xiangya Hospital in neighboring provinces, rather than visiting nearby facilities.

Since 2016, Yuanzhou District in Yichun City has piloted the remote imaging diagnosis model of “grassroots examination, superior diagnosis” at 10 township health centers, including Liaoshi Town Health Center. JF HEALTHCARE, as the technology provider, was deeply involved in the construction of this project.

After patients undergo medical imaging at township health centers, the images are uploaded to JF HEALTHCARE’s system for diagnosis by its medical experts. Within 15–30 minutes, the township health centers receive standardized, high-quality diagnostic reports. During the four-month pilot period, JF HEALTHCARE provided 5,384 remote imaging reports to 10 township health centers, alleviating the difficulties faced by residents in mountainous areas in accessing medical care.

Furthermore, Jiufeng’s remote medical diagnostic services provide a two-way interactive platform for guidance and training, connecting specialists with primary care clinicians. For instance, physicians at township health centers can compare their initial diagnoses with the remote diagnostic results. Should any questions arise, they can engage in timely discussions and communication via the platform, helping to reduce misdiagnoses and missed diagnoses. According to survey data from the Yichun Municipal Committee’s Office of Reform, patient satisfaction with this service has reached 92%.

It is reported that JF HEALTHCARE, established in May 2015, is a medical technology company headquartered in Nanchang. Its R&D team is based in Shanghai and Hangzhou, with an imaging physician studio located in Nanjing. Currently, the company employs 75 staff members, including 20 in operations, 22 physicians, and 25 in research and development.

The company targets primary care hospitals as its entry point, capitalizing on the acute shortage of radiologists in these facilities, where demand is rigid and willingness to pay is strong. It initially launched with DR image interpretation services and has since gradually expanded into other domains, such as ECG and ultrasound.

By establishing third-party medical imaging centers, JF HEALTHCARE provides imaging diagnostic services to primary-care hospitals through a B2B model. Founder Wu Wenhui stated, “JF HEALTHCARE’s application scenario adopts a human–AI collaborative approach: it first performs diagnostics on the imaging data uploaded by hospitals, then uses computer-aided tools to annotate the findings; after physician review and sign-off, JF HEALTHCARE charges a service fee for each report.”

As of 2017, JF HEALTHCARE had signed contracts with more than 500 township health centers, cumulatively interpreting 190,000 images and annotating over 140,000 X-ray films.

Due to the inconsistent quality of images captured by equipment at primary healthcare facilities, Jiufeng has established a workstation to ensure consistency. This workstation performs preprocessing and analysis of X-ray images within 1–2 seconds before they are transmitted from township health centers, informing physicians whether the image quality meets the required standards. If any anatomical region is missing or the patient positioning is incorrect, the system provides alerts, allowing physicians to opt for retaking the images.

According to data from the Jiufeng system, approximately 10%–20% of radiographic images taken by physicians at township health centers are currently of substandard quality. Guidance provided through front-end workstations can ensure overall image quality. In addition to reducing rates of misdiagnosis and missed diagnoses, this approach helps improve the technical proficiency of radiologists at the primary care level.

Furthermore, given the nature of the Jiufeng Third-Party Imaging Center, all X-ray films transmitted back to hospitals through Jiufeng will bear the signatures of corresponding radiologists. In the event of any disputes or liability claims arising from these images at township health centers, Jiufeng will assume full responsibility.

A team led by He Jian, an algorithm scientist at JF HEALTHCARE, participated in a bone medical imaging competition launched by Andrew Ng and the Stanford ML team. Leveraging the MURA dataset released in December 2017, the competition aims to drive significant advancements in medical imaging technology, achieving diagnostic accuracy on par with human experts, thereby enhancing radiologists’ workflow efficiency and improving radiology care standards.

It is reported that musculoskeletal diseases affect 170 million people worldwide, representing the most common cause of severe chronic pain and illness, and resulting in 30 million emergency department visits annually. MURA (Musculoskeletal Radiographs) is a large dataset of bone X-rays and one of the largest open-access radiological imaging datasets globally.

MURA employs a hidden test dataset to officially evaluate model performance. Participating teams can submit executable code on Codalab, which is then run on a non-public test dataset; this process largely ensures the fairness of the test results. Once models undergo official evaluation, their test scores are displayed on the leaderboard.

As of now, the results submitted by JF HEALTHCARE’s algorithm team rank first on the leaderboard, demonstrating performance that is largely comparable to the level of a senior radiologist.

In addition to a top-tier technical team, the ability to achieve outstanding results is also supported by imaging experts. To this end, JF HEALTHCARE has engaged experts such as Professor Chen Junkun and Professor Li Baoxin, Directors of the Department of Radiology at Drum Tower Hospital Affiliated to Nanjing University Medical School, to establish the JF Imaging Annotation Committee. Currently, the committee has more than ten full-time annotation members. They not only participate in annotation work but also design batch reading workflows for the JF Imaging Center, thereby ensuring the quality of image interpretation.

Wu Wenhui stated, “JF HEALTHCARE does not engage physicians to perform annotation tasks through third-party labeling platforms, as this approach not only incurs high costs but also makes it difficult to ensure the quality of annotations.” It is precisely due to the high quality of JF HEALTHCARE’s annotations that its artificial intelligence system achieved superior accuracy in the skeletal medical imaging competition, even when algorithmic performance differences were minimal.

AI + Telemedicine + Imaging = JF HEALTHCARE’s Closed-Loop Service, the World’s First X-ray Imaging Diagnostic Assembly Line.

In response, Wu Wenhui stated: “JF HEALTHCARE is by no means creating this closed loop merely to jump on a buzzword bandwagon. Given that township health centers currently lack the capacity and incentive to purchase auxiliary diagnostic software or hire professional radiologists, JF HEALTHCARE aims to meet their needs by providing both physician services and integrating telemedicine with artificial intelligence. This approach not only reduces costs for township health centers but also enhances operational efficiency for both JF HEALTHCARE and these facilities.”

Currently, JF HEALTHCARE primarily generates revenue by participating in targeted poverty alleviation initiatives led by local governments. Moving forward, the company plans to drive additional revenue streams through the productization of big data and artificial intelligence solutions, as well as by providing specialized data services. For instance, it aims to collaborate with Centers for Disease Control and Prevention (CDCs) at various administrative levels nationwide to leverage big data analytics in generating regional population health reports.

The penetration of information systems in township health centers remains low in China, posing significant challenges for the three national teams currently being established to develop big data in healthcare and medicine. Collecting data from primary care facilities is particularly difficult. However, beyond tertiary (Grade 3A) and secondary (Grade 2A) hospitals, primary care medical data is an indispensable component, making it a highly contested area for major corporations. JF HEALTHCARE’s business aligns precisely with the objectives of these national teams, paving the way for potential collaboration in the realm of big data.