On speech recognition during anaesthesia

Abstract

This PhD thesis in human-computer interfaces (HCI, informatics) studies the case of the anaesthesia record used during medical operations and the possibility to supplement it with speech recognition facilities. Problems and limitations have been identified with the traditional paper-based anaesthesia record, but also with newer electronic versions, in particular ergonomic issues and the fact that anaesthesiologists tend to postpone the registration of the medications and other events during busy periods of anaesthesia, which in turn may lead to gaps and inaccuracies in the anaesthesia record. The thesis first studies the role and the importance of the anaesthesia record as a work tool during operations. Related work procedures are also described in detail. Some small-scale surveys are conducted, which corroborate the observations mentioned above. Supplementing the electronic anaesthesia record interface with speech input facilities is proposed as one possible solution to a part of the problem. A discussion paper made with a socio-ergonomist describes some of the short and long-term consequences if such an idea is to be deployed. The thesis then investigates the possibilities and technical limitations of the most widely used speech recognition system in Danish for medical applications. Of particular interest is the deleterious effect of various background noises found in medical operation theatres. While loud noises in the operating room can have a predominant negative effect, recognition rates for common noises are found to be only slightly below performances obtained in an office environment. Other factors have a major impact as well, such as the words to be recognised, participants, the type of speech recognition system (natural or constrained language) and the type of microphone. Finally, a proposed redundant architecture succeeds in improving the reliability of the recognitions. After that, a prototype of electronic anaesthesia record interface with speech input facilities is developed on the basis of the knowledge gained at the previous steps as well as interviews with some anaesthesiologists. The next phase is based on full-scale anaesthesia simulations involving the prototype, to compare it with the traditional touch-screen and keyboard interface. Inspired from the mathematical queuing theory, a special metric for characterizing differences in mental workload is developed to compare the two interfaces. Results show that the speech interface leads to much shorter registration delays and to a greater accuracy of the information than the traditional electronic interface. 6 Alapetite 2007: On speech recognition during anaesthesia. PhD thesis. The simulation-based experiments also permitted the testing of some speech input strategies chosen for the prototype (hands-free vocal interface activated by a keyword; combination of command and free text modes), which were successful, even with the ambient noise. Speaking to the system while working appeared feasible, although improvements in speech recognition technologies are still necessary. The above experiments form the main results of the thesis. They are followed up by secondary investigations. An opportunity is taken to study via questionnaires and other indicators the deployment, acceptance and success of a speech recognition system – sharing technological similarities with the above-mentioned prototype – used to produce patient records in a Danish hospital. Physician satisfaction with the use of the system is modest, yielding a posteriori an approximately even balance between those in favour of, and those against the introduction of speech recognition to transcribe medical records. One of the main reasons for users’ dissatisfaction is the new work procedure introduced simultaneously with the speech recognition technology, which requires physicians to spend more time on producing the records. In order to get more objective data on the effect of introducing this speech recognition system on the quality of the medical records, a blinded comparison is done between former and new work procedures. The results show that records produced with the new work procedures involving possible use of speech recognition contain more errors than the ones produced with the former method where a secretary is in charge of the transcription. However, the difference between speech recognition and secretary based transcription is relatively small in terms of number of transcription errors, and does not apply to records that follow a fixed and recurrent pattern. These results may therefore not be construed as showing that speech recognition does not bring advantages when considering the gains, e.g. in total turnaround time. The conclusion is that speech recognition is a very interesting modality that should be used when appropriate and only for tasks for which it is efficient when compared to other alternatives.