Click the blue words above to follow us, and click “Write a comment” at the end of the article to express your views.
e-Medical Pang Tao
The application of voice recognition technology at Peking University People’s Hospital (hereinafter referred to as “PKUPH”) began with the Radiology Department. As the first doctor to engage with and assist the information center in implementing this technology, Li Zhentao described the debut of the voice recognition technology provided by Beijing Yunzhisheng Information Technology Co., Ltd. (hereinafter referred to as “Yunzhisheng”) as nothing short of “amazing!” The voice recognition technology began its “journey” in the PKUPH Radiology Department with this halo.
”It is very common for doctors in Europe and America to use voice input; we have seen many successful cases abroad. Especially for radiologists, using voice recognition technology to write reports has become a mature practice.” As the first doctor at PKUPH to engage with and assist in promoting voice recognition technology in the Radiology Department, Li Zhentao is not unfamiliar with this technology and has been looking forward to it for a long time—the process of radiologists writing reports determines the necessity of voice recognition technology.
Radiologists typically write reports on one screen while viewing images on another, frequently switching the mouse focus between the two screens. Therefore, utilizing voice recognition technology can liberate doctors from a large amount of data entry work, allowing them to concentrate all their attention on image diagnosis rather than the burdensome data entry, thus significantly improving their work efficiency. “Many doctors around me have seen and personally experienced the application effects of voice recognition technology abroad, and they are very supportive and eager to have the opportunity to apply it in their own work,” Li Zhentao admitted, radiologists are eagerly anticipating the application of voice recognition technology in their department.
“Amazing” Is Just the Beginning
This technology, which has been the object of radiologists’ dreams, came to the PKUPH Radiology Department through their PACS system provider. The PACS system provider for PKUPH is also the provider for Peking Union Medical College Hospital, and Yunzhisheng’s voice recognition system achieved an accuracy rate of over 95% when applied at Peking Union Medical College Hospital. Based on the successful application there and their understanding of PKUPH Radiology Department’s urgent search for a solution, they introduced Yunzhisheng’s voice recognition system.
Regarding Yunzhisheng’s voice recognition system, Li Zhentao described it as “amazing”: “The accuracy rate is the first and most important standard for evaluating voice recognition systems. We have seen voice recognition systems abroad perform very well with accents, whether it is English with an Indian accent or Chinese-accented English, the recognition rates are very high. So, regarding the accuracy rate that Yunzhisheng’s voice recognition system demonstrated here, everyone was both eager and had a big question mark. Because we heard that it performed very well at Peking Union Medical College Hospital, we approached it with a ‘let’s give it a try’ attitude and had Yunzhisheng install their system.”
The initial trial showed high efficiency and accuracy, a result that Li Zhentao and the radiologists did not expect; Yunzhisheng’s voice recognition system’s “debut” was indeed “amazing,” which also provided an opportunity for further in-depth cooperation between Yunzhisheng and PKUPH Radiology Department.
Exclusive Corpus on a Private Platform
For hospitals, the uniqueness and professionalism of their corpus is a significant characteristic and core requirement. Whether the voice recognition system can be widely and highly applied in hospitals depends on the corpus, which is crucial. For example, in electronic medical records and radiology reports, due to the inconsistency of the input content, each must have its unique corpus to support it to ensure the highest accuracy.
”The successful ‘debut’ of Yunzhisheng is one of the bases for our choice to cooperate with them. What truly made us decide to further cooperate with them is Yunzhisheng’s focus on the medical industry for corpus analysis and corpus development,” Li Zhentao explained: “Previously, we also had voice recognition technology vendors contact us, but after understanding and engaging with them, these vendors were not focused on the medical industry; they were merely interested in it. This means they cannot deeply understand and support the applications in the medical field, which is why we did not choose to pursue further cooperation. Additionally, we also required that the unique corpus be made into a private cloud platform and placed on the hospital’s server. This is also a specific requirement that many vendors with experience in voice recognition technology cannot meet.”
During the deeper cooperation process, PKUPH Radiology Department provided Yunzhisheng with some “de-identified” reports, which Yunzhisheng’s technology engine analyzed to generate a professional corpus exclusive to PKUPH Radiology Department. “This corpus, after being applied, raised the accuracy rate to over 95%. Moreover, from the collection of report samples to the completion of corpus design, Yunzhisheng took less than two weeks.” The stable and secure application is the core idea of applying any information system, and Yunzhisheng’s response has truly satisfied them.
The structured reports of PKUPH Radiology Department’s PACS have long been praised in the industry. After practical application, Li Zhentao stated that the efficiency of input using the voice recognition system combined with report templates is also very high, providing a good standardization process. The integration with the existing information system fills PKUPH Radiology Department with confidence in maximizing the effectiveness of this system.
Meticulous Details Make Application More Natural
After successfully addressing efficiency and accuracy issues, the collaboration between PKUPH Radiology Department and Yunzhisheng upgraded to the level of “how to meet the actual needs of doctors in detail,” allowing this system to evolve from being usable to being user-friendly. “The standard for a user-friendly system must meet the needs of doctors. For example, the issues of voice input and punctuation recognition illustrate this well,” Li Zhentao said.
In a report, there are two roles involved: the reporting physician and the auditing physician.
For the reporting physician, writing radiology reports primarily consists of short sentences, unlike electronic medical records which are in long sentence form. However, the initial design of Yunzhisheng’s voice recognition system defaulted to recognizing automatic punctuation, which was not suitable for the reporting physician’s input method. It was more in line with the doctor’s needs to read the punctuation aloud and have it recognized and completed. For the auditing physician, they do not need extensive input but only need to modify partial errors or descriptions of certain words, so they do not require automatic punctuation recognition either.
Based on these practical scenarios, Yunzhisheng modified its system, and such detailed modifications significantly enhance user experience.
The structured report system templates of hospitals are often the crystallization of the collective wisdom of departments, and each doctor needs to undergo a process of understanding the report templates. How to enable the voice recognition system to help doctors better understand the templates, avoiding input issues arising from misunderstandings of the templates, is also one of the indicators of whether this system is “user-friendly”.
”The hierarchy of structured reports is extensive, like a large tree, from the trunk to several branches, and then to branches on branches… Each level is extremely important,” Li Zhentao detailed how to evaluate the “polished” system. During the input process of structured reports, doctors are more like filling in blanks. Through the voice recognition system, it can intelligently analyze the key points of input, and when doctors speak, it automatically analyzes which can correspond to these key points, which is the most convenient for doctors. It is also quite natural for doctors to make deletions and modifications during input; how to allow doctors to quickly operate the interface while deleting and modifying according to the hierarchy without disrupting the system’s default report structure is also a specific performance of whether the system is user-friendly.
A system that better aligns with doctors’ usage habits is not only reflected in the system itself; in the design of microphone shortcut functions, PKUPH Radiology Department and Yunzhisheng also conducted in-depth research based on practical scenarios. “Pressing starts input, releasing stops input; one-click delete, backspace. These can all be customized easily to fit doctors’ actual operating scenes and habits,” Li Zhentao expressed great satisfaction with such flexible customization of function keys, and he also recognized the input effect: “Philips’ microphone performs excellently in noise reduction and directional recording; even with several doctors sitting in close proximity, recognition poses no issues.”
Respecting and Changing Habits
At the beginning of the application of the voice recognition system in the department, many doctors had a common concern: if my Mandarin is not very standard, will the recognition rate be affected, thus impacting efficiency? PKUPH Radiology Department gathers excellent medical talents from all over the country, along with some doctors coming from other hospitals for further training. Recognition of dialects is indeed a realistic and tricky problem in everyone’s view. Out of respect for users’ language habits and to better enhance the system’s usability, Yunzhisheng’s voice recognition system is designed with strong language adaptability, capable of automatically improving recognition efficiency and accuracy based on the user’s speech habits after repeated use.
”We invited colleagues from representative regions in the department to use this system, and after multiple uses, the recognition efficiency and accuracy significantly improved, reaching the standard Mandarin recognition capability,” Li Zhentao explained.
Also concerning “habits,” Yunzhisheng brings change and optimization possibilities for doctors’ futures. Li Zhentao noted that the reports written by radiologists are for clinical doctors, and how to improve the coordination between the two requires thoughtful research and change. The voice recognition system makes input smarter, so finding a point of alignment between text input and structuring, allowing radiologists to focus solely on the diagnostic content while the support for structuring is handled by Yunzhisheng’s system, and making those relatively complex and lengthy reports appear and read more quickly and directly to support diagnosis, is also one of the challenges Yunzhisheng will continue to tackle in the future!
e-Medical August 2016Monthly Edition
*** For subscription methods to “Technology New Era” e-Medical, please refer to the menu bar of this public account “Us – Magazine Subscription”
***Except for all copyright images in this journal, the remaining images in this article are sourced from 360 images. If there are any copyright issues, please contact us for deletion.
© The above is sourced from
e-Medical August 2016 original article, please indicate the source when reprinting.
