Andi Medical Intelligence Secures China's First Class III NMPA Approval for AI-powered MR-based Intracranial Tumor Diagnostic Software

Following the Major Test of the Pandemic, China’s First Class III Certificate in the Field of AI-Assisted Imaging Diagnosis Achieves a Breakthrough. On June 12, the National Medical Products Administration (NMPA) announced that BioMind’s “Tianyi Zhi” MRI-Assisted Diagnostic Software for Intracranial Tumors has passed the NMPA Class III medical device approval.

This product was developed by the "Artificial Intelligence Research Center for Neurological Diseases," jointly established by Beijing Tiantan Hospital and BioMind.It is the first medical AI software approved by the National Medical Products Administration (NMPA) under the name “Image-Assisted Diagnosis.”

It is reported that BioMind, headquartered in Beijing, has established wholly-owned subsidiaries in Singapore and Luxembourg. This approval marks the first Class III medical device certification in China for AI-assisted diagnostic imaging based on deep learning technology.

The market has been dormant for a long time. How has BioMind managed to break through at this juncture? What implications does the approval of its AI product for intracranial tumors hold for the industry?

The product approved this time is the AI-assisted diagnostic software for intracranial tumors in MR, part of BioMind’s “Tianyi Zhi” series. Leveraging deep learning algorithms trained on hundreds of thousands of pathology-confirmed brain tumor cases, it enables precise AI-based diagnosis of intracranial tumors (such as meningiomas, vestibular schwannomas, medulloblastomas, and gliomas), with an overall diagnostic accuracy exceeding 90%, and surpassing 96% for certain tumor types. The system automatically generates a structured report containing precise information such as tumor location and volume, thereby assisting physicians in making rapid diagnoses and enhancing radiologists’ diagnostic capabilities for brain tumors.

When discussing the decision to focus R&D efforts on intracranial tumors, Wu Zhenzhou, Chief Technology Officer of BioMind, recalled the rationale behind the choice: “It was primarily driven by considerations of market size and technical complexity. Since MRI involves multi-sequence imaging, it presents significant analytical challenges, compounded by the intricate nature of intracranial anatomy. Very few companies in China are dedicated to MRI-based analysis of intracranial tumors, which presented us with a strategic opportunity.”

From a market perspective, the intensification of urbanization, industrialization, population aging, and globalization, coupled with environmental degradation, changes in lifestyle, and the influence of biological and genetic factors, has led to a continuous increase in both the frequency and level of exposure to risk factors for malignant tumors. Consequently, the global incidence and mortality rates of malignant tumors are on the rise, posing a serious threat to human health and socioeconomic development.

Imaging of central nervous system tumors exhibits the phenomena of “different diseases with similar imaging features” and “the same disease with varying imaging presentations.” Therefore, diagnosis cannot rely solely on imaging findings; it must be based on a comprehensive assessment that includes the patient’s medical history, clinical symptoms, physical signs, and other relevant auxiliary examinations. This implies that a qualified medical imaging professional must not only master knowledge of imaging diagnosis but also understand and grasp a range of clinical knowledge, including disease etiology, pathology, clinical symptoms, physical signs, laboratory test indicators, and even treatment principles. Furthermore, they must develop rigorous logical thinking skills to comprehensively analyze various morphological and functional imaging features, weigh multiple factors, and make integrated judgments, thereby gradually improving the accuracy and comprehensiveness of brain tumor diagnoses. Such an extensive process of accumulation and learning results in a prolonged training period for excellent imaging professionals, leading to a growth in the number of radiologists that falls far short of meeting the ever-increasing workload demands in diagnostic services.

Therefore, the core issue BioMind aims to address is the scarcity of medical resources. By leveraging AI-assisted diagnosis, BioMind empowers hospitals and physicians, helping to mitigate the severe imbalance in diagnostic and treatment capabilities. As the sole medical AI enterprise representative invited to appear on CCTV’s “Dialogue” program, BioMind conveyed its social mission to the Chinese people: medical artificial intelligence can serve as a tool for decentralizing high-quality medical resources, enhancing the clinical proficiency of healthcare providers at primary care institutions, and enabling patients to access high-standard diagnostic and treatment services locally. This represents the true significance of medical artificial intelligence.

According to Li Jingjue, CEO of BioMind China, “The ‘intelligence’ and ‘capability’ in medical AI products should carry distinct connotations compared to AI in other fields. ‘Intelligence’ should embody the wisdom and clinical expertise of top-tier specialists from leading hospitals. Input determines output—this is fundamental. ‘Capability’ should represent empowering clinicians, particularly by enhancing the diagnostic and treatment proficiency of physicians in primary care hospitals. Through close collaboration with Beijing Tiantan Hospital, BioMind’s ‘Tianyi Zhi’ leverages deep learning algorithmic models, systematically trained on over one million tumor imaging cases from the past decade, while integrating the clinical experience of experts from premier hospitals.”

In addition to the aforementioned approved AI-assisted diagnostic software, BioMind’s “Tianyi Zhi” product series also covers precise diagnosis of multiple organs and diseases in the head, neck, heart, blood vessels, and breasts. It is committed to becoming an AI-assisted diagnostic solution for CT and MR imaging covering the entire body, enabling functions such as AI-assisted diagnostic alerts and risk assessment, and deeply integrating into core hospital medical service workflows. It is reported that other products in the BioMind “Tianyi Zhi” series are currently undergoing clinical trials and applying for Class III medical device registration.

“AI applications based on single scenarios, such as a single anatomical site or a single disease, have limited clinical value and represent merely a transitional phase in the development of medical AI. The R&D of ‘Tianyi Zhi’ began with neurological disorders of the head, then expanded to auxiliary diagnosis for diseases of the heart, breast, and cardiovascular system, as well as other thoracic and abdominal conditions. Achieving AI applications across multiple anatomical sites and disease types is our overall future product R&D direction,” Li Jingjue stated in a previous media interview.

This systematic and process-driven product model is grounded in real-world application scenarios for common, frequently occurring, and acute or critical conditions, thereby genuinely meeting the daily needs of physicians. Furthermore, it goes beyond merely assisting physicians in qualitative disease diagnosis; more importantly, it enables precise risk assessment of disease progression and supports clinical decision-making.

Quoting Paul Chang, Chair of Radiology at the University of Chicago Medical Center, at the 2019 RSNA Annual Meeting: “We only truly begin to understand AI when we turn our attention to clinical needs, applications, and value.” In this regard, Li Jingjue stated that BioMind’s “Tianyi Zhi” platform, with risk assessment as its core and process optimization as its hallmark, can be applied across multiple hospital departments—including diagnostic radiology, emergency medicine, neurology, neurosurgery, neurointervention, cardiology, and breast surgery—thereby meeting diverse clinical needs.

Perhaps, by continuously embedding AI into integrated diagnosis and treatment through iterative updates, BioMind is delivering the best practice of the aforementioned essence of AI.

So, why was BioMind’s AI diagnostic product for intracranial tumors able to break through the competition and gain approval? VCBeat has compiled the interview content in an attempt to uncover the logic behind the approval.

“Regarding the approval of medical AI, the professionalism and efficiency demonstrated by officials from the Drug Registration Department and the Center for Medical Device Evaluation (CMDE) under the National Medical Products Administration (NMPA) were truly impressive,” Li Jingjue told VCBeat.

In fact, many AI-powered medical products have been delayed in obtaining regulatory approval, likely due to bottlenecks in the clinical trial phase. It is reported that many medical AI companies have not conducted effective clinical trials for their products and have not yet entered the approval process.

It is precisely due to the clinical trial phase that the product fails to demonstrate non-inferiority and superiority.

When evaluating the efficacy of clinical trials, non-inferiority trials are designed to test whether one drug is not inferior to another. These trials are commonly used in clinical studies with objective efficacy endpoints, such as clinical outcomes for antimicrobial agents, adverse events in cardiovascular therapy, and death or disease progression in oncology treatment. In contrast, superiority trials are designed to test whether one drug is superior to another; this design is generally employed in trials using a placebo as the control.

In the regulatory approval of medical artificial intelligence devices, although the definitions of non-inferiority and superiority differ in form, they are similar in essence. To demonstrate either superiority or non-inferiority, an AI product must prove that it has clinical application value or offers advantages over existing products. Therefore, clinical trial design undoubtedly constitutes a critical step in securing regulatory approval.

Another issue lies in the source of data. During the early stages of the rise of artificial intelligence, companies obtained medical data through various channels and used it for AI training. However, with the release of the “Review Points for Medical Device Software Assisted by Deep Learning” and the “Appendix to the Good Manufacturing Practice for Medical Devices: Standalone Software,” along with the increasing standardization of cybersecurity regulations, companies are now required to explain the valid sources of their training data to the Center for Medical Device Evaluation (CMDE) during the approval process. This has forced some companies to retrain their AI products.

In this light, BioMind’s “Tianyi Zhi” MRI-assisted diagnostic software for intracranial tumors, which has recently received regulatory approval, demonstrates a robust practical approach by Ande Medical Intelligence. Rather than being confined to early disease screening, the software enables further precise diagnosis, graded alerts, and risk assessment, thereby delivering genuine clinical value by actively participating in diagnostic and therapeutic decision-making.

So, how should enterprises address the numerous issues mentioned above? And what does the future hold for their development? Drawing from BioMind’s experience, we can distill three strategic pathways.

For a long time, we have referred to the application of artificial intelligence in healthcare as “AI + Healthcare,” a term that seemingly positions AI as the primary driver propelling the development of the medical sector. In practice, however, as AI is increasingly adopted across various clinical scenarios without becoming their sole dependency, it has become evident that specific use cases dictate AI implementation. AI serves to empower existing healthcare workflows—making “Healthcare + AI” the more appropriate and sustainable paradigm.

Returning to medical imaging, many existing imaging devices—such as CT, MRI, color Doppler ultrasound, ECG, EEG, and X-ray—have incorporated artificial intelligence to varying degrees. However, for AI to truly deliver value, companies must avoid the trap of equating a single feature with a standalone product.

For example, when a patient presents with fever and headache, physicians cannot immediately determine the specific underlying condition. If the patient undergoes an MRI, a single-function product—such as one designed solely for detecting intracranial hemorrhage—fails to meet clinical needs. What physicians require is an artificial intelligence system capable of comprehensively assessing the patient’s cerebral status. In industry terminology, this translates to AI products that provide “full-spectrum disease coverage” for at least a specific anatomical region.

This represents a development trend and is one of the optional pathways for enterprises designing clinical trials. Based on current observations, medical AI companies led by BioMind have chosen and implemented this direction—striving to develop products capable of diagnosing multiple anatomical sites and a wide range of diseases.

Once a suitable scenario is identified, companies need to leverage data to refine their products, but acquiring such data is no simple task.

From the perspective of existing algorithmic mechanisms, if AI products are trained using effective data from primary healthcare settings, their performance ceiling will be limited to general applications within primary care, rendering them incapable of extending to large-scale hospitals.

“For the diagnosis of diseases such as breast cancer and brain tumors, there are significant disparities across hospitals at different tiers. If data are selected arbitrarily, increased training may well lead to decreased accuracy,” stated Wu Zhenzhou.

Therefore, for medical AI products to be successfully implemented in Grade A tertiary hospitals, they must leverage high-quality data from top-tier hospitals and deeply learn the “gold standard” clinical expertise of leading specialists to ensure the accuracy of the AI.

For a long time, the barriers to entry for AI in healthcare may not have been so apparent—companies could rapidly catch up and surpass incumbents simply by gaining access to high-quality data. Today, that has all changed.

“Many AI companies will soon discover that previously open-source algorithms are becoming less effective. As we gradually advance toward comprehensive coverage of all disease types, single-task deep learning algorithms can no longer meet the demand; multi-task algorithms represent the prevailing trend,” said Wu Zhenzhou. Since its inception, BioMind has assembled a team of elite researchers, all hailing from top institutions in the field of deep learning such as Harvard University, MIT, the National University of Singapore, Tsinghua University, and the Chinese Academy of Sciences. The team specializes particularly in the simultaneous analysis of multiple diseases.

Therefore, in addition to continuing the race for high-quality, effective AI data, medical AI companies must seek breakthroughs at the algorithmic level in the next phase.

Overall, 2020 marked a promising start with the approval of BioMind’s Class III medical device certification for its AI-powered medical tools. However, the path forward for medical artificial intelligence remains long. A broad view of the current medical imaging AI market reveals that competition among companies is hardly even a factor—few are capable of delivering mature AI products.

Therefore, while the approval of Class III medical devices—representing the collective efforts of numerous AI in healthcare practitioners—is undoubtedly important, it is even more crucial for companies to continue refining their algorithms, securing data resources, and staying grounded in clinical practice, so as to fundamentally reshape the current landscape of AI in healthcare. BioMind, having already obtained its Class III medical device certification, will continue to accelerate its progress in this field.

As BioMind stated, “The essence of medical AI’s value lies in addressing clinical needs. How to center future strategies on these needs and leverage AI technology to empower healthcare has always been our foothold in the industry and remains the core mission we must steadfastly uphold.” The journey for medical artificial intelligence is long and arduous, yet thanks to the efforts of numerous medical AI practitioners, including BioMind, a glimmer of hope has emerged, promising a bright future.