Exploration of Mobile Electronic Medical Records Based on Voice Recognition

With the rapid development of the internet and the widespread use of mobile terminals, voice recognition technology has made significant advancements with the help of artificial intelligence, data mining, and other information technologies. Particularly in the medical field, voice recognition technology has been widely adopted in hospitals in Europe and the United States, primarily for medical documentation entry, saving doctors’ input time and allowing more time for communication with patients. In recent years, the application of voice recognition in Chinese hospitals has also been deeply explored and researched. However, due to factors such as doctors’ usage habits and input recognition rates, it has not been widely implemented. On the other hand, the diagnosis and treatment behaviors centered on electronic medical records remain a key business for hospitals, and effectively reducing the information entry workload of clinical doctors remains an urgent issue to be solved. Against this backdrop, there is a need to better apply voice recognition technology in the electronic medical record information entry process to construct higher quality and more efficient electronic medical records.

Voice recognition is a technology that allows machines to convert speech signals into corresponding text or commands through recognition and understanding processes. It enables machines to understand human speech, converting the words spoken by users into text word by word, and displaying the text accurately. Voice recognition is an interdisciplinary field, involving physiology, acoustics, linguistics, computer science, signal processing, etc. Although different voice recognition systems have varying implementation details, the underlying principles are generally the same, as shown in Figure 1.

Figure 1 Voice Recognition Principles

Voice recognition technology is mainly divided into two parts: front-end processing and back-end processing. Front-end processing includes endpoint detection, noise reduction, and feature extraction. Back-end processing consists of two processes: training and decoding.

Endpoint Detection detects valid speech segments from continuous speech streams. It involves two aspects: detecting the starting point of valid speech (the front endpoint) and detecting the ending point of valid speech (the back endpoint).

Noise Reduction The audio collected in practice generally has background noise. When the noise intensity is high, it can lead to a decrease in voice recognition rates and sensitivity in endpoint detection. Therefore, noise suppression is necessary.

Feature Extraction involves extracting and selecting acoustic features. Common methods include Linear Predictive Coding (LPC), Cepstral Coefficients (CEP), and Mel Frequency Cepstral Coefficients (MFCC) calculations.

Training involves pre-analyzing speech feature parameters, creating speech templates, and storing them in a speech database.

Decoding involves analyzing the speech to be recognized in the same way as during training to obtain speech parameters, comparing them with reference templates in the database, and using decision-making methods to find the template that most closely matches the speech features to achieve recognition results.

Needs Analysis With the continuous deepening of hospital information construction and the ongoing improvement of clinical information system implementation, most doctors’ daily work has become reliant on the support of hospital information systems. Particularly, the entry of medical documentation information has become a routine task. Research indicates that over 40% of doctors spend about 4 hours a day entering text in front of computers, accounting for 40% of their daily work time. It is evident that the convenience of text entry has become a significant factor affecting doctors’ work efficiency. Although the computer-based electronic medical record entry provides some template references, the efficiency of information entry cannot be effectively improved due to individual patient conditions and the necessity to be at the computer. With the widespread use of mobile smart terminals, mobile electronic medical records effectively solve this problem, allowing clinical doctors to combine speaking and writing in fragmented time, enabling them to enter electronic medical record information anytime and anywhere, significantly improving the efficiency of information entry.

System Architecture The architecture of the voice recognition-based mobile electronic medical record application is mainly divided into two parts. The first part is the internal network servers deployed within the hospital, including the internal mobile doctor workstation proxy server and electronic medical record server. The second part is the medical voice cloud server deployed on the external network. To ensure secure communication between the internal and external network servers, an SSL VPN connection is used. The system architecture is shown in Figure 2.

Figure 2 System Architecture Diagram

Functional Design Clinical doctors install the mobile doctor workstation on their smartphones and utilize the mobile electronic medical record entry function to complete various electronic medical records’ writing, seamlessly connecting with the computer-based electronic medical records for bidirectional synchronization, forming a complete electronic medical record management closed-loop process. This article mainly introduces the text entry process based on voice recognition, as shown in Figure 3.

Figure 3 Mobile Electronic Medical Record Entry Process

Medical record information entry: Doctors use the phone’s speaker function to input speech, or they can use the phone’s keyboard for traditional text entry. Speech transcription and semantic analysis: According to the principles of voice recognition, speech is converted into text and stored and displayed. Speech input confirmation: The text automatically recognized by the system is reviewed by the doctor for completeness, spelling errors, and necessary modifications. Information synchronization: Electronic medical records added, modified, or deleted on the mobile side must ultimately be synchronized to the computer side to maintain the integrity of electronic medical records. The computer can see the information operated on the mobile side in real-time. Any additions, deletions, or modifications made by the doctor on the computer side will also be synchronized to the mobile electronic medical records in real-time. Signature and submission: Since the mobile side has not resolved the issue of doctor signatures, all electronic medical records entered on the mobile side must be signed on the computer side before submission.

Currently, the mobile electronic medical record function has been applied in all clinical departments of Wuhan Central Hospital. After six months of practical application, the voice recognition-based mobile electronic medical record entry function has gained widespread recognition from hospital doctors, who generally believe it can quickly convert speech into text and allows information entry operations anytime and anywhere, thus saving text information entry time and improving work efficiency. At the same time, the saved time can be used for doctor-patient communication, increasing patient satisfaction. According to statistics, in February 2018, over 600 electronic medical records were entered using mobile terminals, accounting for nearly 10% of the total electronic medical records in the hospital, with doctors saving an average of about one hour of text information entry time per day.

Based on long-term application exploration and user feedback, it was found that the promotion of voice recognition-based mobile electronic medical records in clinical departments faces several issues, including doctors’ unfamiliarity with operations, privacy concerns over verbal descriptions of medical conditions, and the following problems.

Electronic Signature Issues To facilitate clinical diagnosis and treatment work, all information systems on the hospital’s computer side that involve clinical doctor signatures have adopted electronic signatures, completely replacing handwritten signatures. Due to issues regarding the legality, validity, and technical implementation of electronic signatures on mobile smart terminals, the hospital’s mobile information systems have not launched electronic signature functions, requiring doctors to return to the computer side to electronically sign the medical record information entered on the mobile side before successful submission. This adds an extra step to the clinical doctors’ mobile electronic medical record entry process, severely limiting their enthusiasm for inputting information. Therefore, the hospital began exploring the application of electronic signatures on mobile smart terminals at the end of 2017.

Voice Recognition Rate Issues Although the accuracy of voice recognition can reach over 95%, the noisy working environment of doctors and varying accents can lead to issues such as extra words, missing words, and recognition errors during the voice recognition process, thereby reducing the accuracy of voice recognition. This is also a significant factor affecting the promotion of voice input. To address this issue, users have organized and customized voice models for the medical field, utilizing a professional medical voice database. During voice input, the system prioritizes the recognition of clinical terminology to reduce the error rate. Moreover, the voice recognition system continues to improve and has added channels for recognizing regional accents for different clinical doctors to choose from.

Information Security Issues Due to the convenience of carrying mobile smart terminals and the need for internet access in the application environment of mobile electronic medical records, the risk of patient information leakage has increased. In addition to using SSL VPN for communication over the internet, measures such as firewalls and intrusion detection are employed to prevent patient information from being tampered with or leaked during communication. Furthermore, the information system has also implemented strict access control measures, allowing each clinical doctor to operate only on electronic medical records of patients within their authorized department. The system will automatically log out if there is no operation within 30 minutes of logging in, thereby protecting patient information security.

Source: “China Digital Medicine” Magazine, Issue 04, 2018, Authors and Affiliation: Liu Jing, Luo Jincheng, Zuo Xiuran, Information Center, Wuhan Central Hospital.