Real-World Feedback on Medical AI: Perspectives from Seven Types of Physicians on Its Clinical Value

Since gaining prominence in the second half of 2016, medical artificial intelligence (AI) has seen over a year of intensive development. Many enterprises’ AI products are now undergoing clinical trials, while some have already been commercialized and are being sold as companies explore revenue models. What is the actual impact of AI products on physicians after they begin interacting with them? Furthermore, what are the attitudes of physicians at different career stages toward AI after using it?

VCBeat, with curiosityInterviewed core medical personnel at various stages, including directors of primary-care hospitals, Party secretaries of Grade 3A hospitals, oncology experts, and radiologists., and gather insights from other physicians, aiming to interpret the current state of medical artificial intelligence from a clinician’s perspective through their representative viewpoints.

On April 12, the Staff Hospital of the 55th Research Institute of China Electronics Technology Group Corporation officially launched the “DE-Ultrasound Robot” AI-assisted diagnostic project for thyroid cancer diagnosis. This project was jointly developed by Huiyi Online and scientists from Zhejiang Deshang Yunxing Imaging Technology Co., Ltd. It is reported that this ultrasound robot has previously been applied in clinical practice at the First Affiliated Hospital of Zhejiang University, achieving a diagnostic accuracy rate of approximately 86%, and can serve as an auxiliary tool for physicians’ diagnoses.

VCBeat conducted an exclusive interview with Zhang Jianyan, the president of the hospital, to gain insights into how leaders of primary care hospitals perceive AI products.

Thyroid ultrasound has become the preferred method for characterizing nodules, owing to its prominent advantages such as high resolution for soft tissues, non-invasiveness, low cost, and ease of operation.

Zhang Jianyan told VCBeat that the introduction of the “DE-Ultrasound Robot” serves two purposes: first, to assist the hospital’s ultrasound imaging physicians; and second, in response to a recommendation from the medical consortium to which the hospital belongs. Deploying ultrasound robots with diagnostic capabilities comparable to those of associate chief physicians at primary care institutions within the consortium can help enhance these facilities’ initial diagnostic capacity and quality, enabling patients to receive high-quality diagnostic services without having to travel long distances.

Previously, their medical workflow was as follows: outpatient physicians would refer patients with suspected thyroid cancer for ultrasound imaging. Based on the imaging findings, the location of the nodules was identified, and clinical judgment was applied to determine whether the nodules were benign or malignant. If a definitive conclusion could not be reached, pathological examination was performed.

However, thyroid cancer often presents without obvious clinical symptoms, and its diagnostic indicators are complex with nonspecific features. Therefore, diagnosis requires a comprehensive analysis integrating the patient’s age, sex, medical history, physical signs, and various test results, which places high demands on physicians’ clinical experience.

Now they have introduced the “DE-Ultrasound Robot.” After the patient’s imaging data is generated,Within five minutes, the robotic system provides an accuracy assessment for thyroid nodules. Physicians then integrate this output with their own clinical judgment to reach a final diagnosis, thereby enhancing diagnostic accuracy. This approach has proven effective based on data from over 100 patients diagnosed to date.

Speaking of the changes brought by artificial intelligence, Zhang Jianyan told VCBeat that in addition to improving the accuracy of doctors' judgments, the most important thing is to enhance doctors' confidence.An AI assistant with the expertise of an associate chief physician is available to help.，Doctors are more confident and assertive when making diagnostic conclusions.

However, in practical applications, artificial intelligence serves merely as an assistant. When conflicts arise between human clinicians and the system—for instance, when the AI identifies a nodule while the physician disagrees—the physician will make the final decision after carefully considering the patient’s specific clinical context.

For the hospital as a whole, the introduction of the “DE-Ultrasound Robot” has indeed led to an increase in patient visits. According to Zhang Jianyan, current patient feedback indicates strong acceptance of this new technology, with patients showing great enthusiasm for AI-assisted diagnosis. No instances of patient resistance have been observed so far.

Regarding the Limitations of "DE-Ultrasound Robot"Zhang Jianyan told VCBeat that the main issue is that the “DE-Ultrasound Robot” only covers thyroid cancer, which is somewhat limited and cannot fully meet hospital needs. Additionally, there are discrepancies with physicians in determining whether a lesion is a nodule.

Finally, regarding the question of whether artificial intelligence could replace doctors, Zhang Jianyan stated that it is still too early for such a development. Doctors at her hospital do not resist AI; currently, AI serves merely as an assistant to physicians, with final clinical decisions remaining in the hands of doctors.

Qingdao Municipal Hospital officially launched Watson for Oncology in April this year. Within just one week of implementation, the hospital has already provided services to more than 10 patients. In light of this, VCBeat interviewed Ding Huamin, Secretary of the Hospital’s Party Committee, to gain his perspective on artificial intelligence.

Ding Huamin told VCBeat that China’s healthcare landscape is characterized by a concentration of top-tier hospitals in Beijing, Shanghai, and Guangzhou, with medical resources in each province primarily clustered in their respective provincial capitals. How do local officials and wealthy entrepreneurs in Qingdao handle cancer diagnoses? Typically, they undergo an initial diagnosis locally, then travel to Beijing or Shanghai for expert consultations before receiving treatment. In contrast, it is particularly difficult for ordinary citizens to secure expert consultations in Beijing or Shanghai, first due to financial constraints, and second because of the challenge in accessing renowned specialists.

After Watson for Oncology was implemented at the hospital, Ding Huamin told VCBeat that major media outlets in Qingdao covered the news. Since then, appointments in the hospital’s oncology department have increased, including among government officials and entrepreneurs. Some of these individuals were already undergoing treatment; they simply sought to use Watson for Oncology to verify whether their current treatment plans were reliable.

There was also an 80-year-old retired worker who, after undergoing surgery for gastric cancer, was advised by his doctor to return for a follow-up examination in three months based on test results. However, having some doubts, he brought newspaper clippings to the hospital to consult Watson. VCBeat has learned that currently, patients scheduling appointments at Qingdao Municipal Hospital are not referred by physicians but instead book consultations on their own initiative.

Qingdao is currently experiencing rapid development and boasts a favorable environment. This has not only attracted residents to settle there but also prompted many hospitals to establish branch campuses in the city. The increased number of hospitals has intensified market competition. Since the introduction of Watson for Oncology, these hospitals have seen a significant rise in patient appointments, which is highly beneficial for enhancing their institutional reputation and revenue.

VCBeat previously compiled and edited Zhu Ying’s speech, in which she stated that “50% of physicians will be replaced by artificial intelligence in the future.” She believes that a certain proportion of tasks performed by junior physicians, as well as those in radiology and pathology, will more likely be assisted by AI-driven systems. This will allow physicians to devote more time to patient communication and clinical research. For details, please click on 《Zhu Ying, Director of the Peking Union Medical College Hospital-affiliated Physician Group: 50% of Physicians Will Be Replaced by AI Doctors》for more details.

Meanwhile, Zhu Ying also stated, for physicians at this stage of their career, the challenge often lies in decision-making. One of her patients was found to have a nodule during examination, and it is currently unclear whether it is malignant. There are two treatment options: one is for the patient to undergo follow-up examinations every three months, and the other is to proceed with immediate surgery. The first option is the safest, but the patient would then have to live each day with the uncertainty of whether they areAmid the shadow of the tumor, the second option is the fastest but carries surgical risks. Should the patient take this risk?

Zhu Ying still struggles to make decisions decisively and hopes that artificial intelligence can assist her in this regard.

Dr. Lindsey Bordone is an attending physician at the Columbia University Department of Dermatology clinic. She sees approximately 50 patients per day, evaluating their conditions and making diagnoses for each. Over her career, she will have managed roughly 200,000 cases. Dr. Bordone stated, “If artificial intelligence can help me make more accurate judgments, I would welcome it. Some of my patients can take photos of their skin concerns before seeing me, which would expand the clinic’s reach.”

However, artificial intelligence cannot replace the communication between doctors and patients. In foreign countries, prolonged doctor-patient interactions help patients relax and feel secure. Meanwhile, dermatologists can perform physical examinations through touch to minimize the risk of misdiagnosis.

After using Infervision’s products, Liu Kai told VCBeat that artificial intelligence can indeed assist radiologists in improving diagnostic efficiency, while also requiring coordinated collaboration across multiple links, including medical equipment and technical personnel.

He also noted that physicians widely believe it is currently impossible for AI to replace them, whereas its role in assisting doctors is tangible and real. Taking a more extreme view, even if replacement were to occur, it would simply reflect societal progress, much like electric lamps replaced kerosene lamps and automobiles replaced horse-drawn carriages. After automobiles supplanted horse-drawn carriages, coachmen transitioned to becoming drivers. Similarly, should physicians ever be replaced by AI, they could still work hand in hand with artificial intelligence.

Dr. Liu stated that many of the current reports can be described as based on false premises. Professionals in related industries, such as enterprises and media, should engage more in conversations with users and doctors, rather than making assumptions off the top of their heads. For more stories about Dr. Liu Kai, please click “Infervision’s Products Deployed at Shanghai Changzheng Hospital; Radiologists Cite Reduced Missed Diagnoses as Their Greatest Benefit》

Gu Xidong was among the first in China to adopt Watson for Oncology and one of the initial partner clients of Hangzhou Cognitive Care. In an interview with VCBeat, Gu stated that the benefits of Watson for Oncology are indeed tangible, such as assisting in the training of young physicians, selecting optimal treatment plans based on evidence-based research, and reducing diagnostic errors by doctors.

However, Gu Xidong also pointed out that artificial intelligence has certain limitations. For instance, Watson is positioned as an assistant to physicians and cannot adjust to the real-life circumstances of patients; it can only recommend treatment plans based on objective pathological indicators. Yet oncology treatment is highly complex: the optimal medical regimen is not always one that patients can accept. In many cases, physicians must tailor treatments to the patient’s actual condition and provide persuasion and emotional support—tasks that Watson cannot perform.

In the episode of the program “Approaching Science” aired on February 14, 2017, Gong Xiangyang’s views on artificial intelligence were presented. The radiology department at his hospital reviews imaging studies for more than 200 patients per day, with each patient having hundreds of images. For any normal human reader, reviewing such a large volume of images inevitably leads to fatigue, which can cause emotional instability. Since image interpretation is a meticulous task, suboptimal mental and physical states increase the risk of missed or incorrect diagnoses.

With the integration of artificial intelligence, AI can perform initial screening to detect and characterize lesions. By comparing AI-generated findings with those of physicians, only cases with discrepancies require expert review. This approach effectively reduces the workload of specialists while improving diagnostic efficiency in image interpretation, without compromising accuracy.

From grassroots hospital directors to hospital party secretaries, from radiologists to oncology experts, these interviews yield conclusions regarding the advantages and limitations of artificial intelligence.

Advantages:

Shortcomings: