Design And Implementation Of Oral Medical Records Recognition System Based On Keywords Automatic Extraction

The applicationof speech recognition technology has begun to gradually expand in the field of medical information industry.It can effectively improve the input efficiency of medical records and reduce the workload of doctors in the electronic medical record system At present,there are two main problems in the electronic medical record system based on speech recognition:First,Spontaneous speech with more Filled Pauses,Word Repetitions,Sentence Restarts is more difficult to be recognized in daily use.Second,the medical recordsfrom the recognition of doctor's oral recordingsare lack of necessary medical text formatleading to that the readability and legibility of the recognitionis not good.To solve the problems above,a spontaneous speech recognition system for medical environmenthas been designed and implemented in this paper;An improved algorithm of keywords extraction based on medical record text has been proposed,which has improved readability and legibility of the medical recordsfrom the recognition of doctor's oral recordings;Finally,The EMR system based on spontaneous speech recognition and keywords automatic extraction has been designed and implemented.The innovations of this paper are as following: the establishment of FP detection model and professional speech recognition system improved the performance of spontaneous speech recognition in medical environment;An improved keywords extraction algorithm based on medical record text has been proposed,which is from the improvement of the term frequency-inverse document frequency algorithm,The new algorithm revises the weight of feature terms from the point of view of the word location,distribution of keyword's part of speech,medical record text classification,and the keywords extracted can reflect topic and key content of medical records better.The main contents of this paper include:(1)Completed modeling and training of FP detection model which is based on GMM-MLP and achieved the FP detection function for spontaneous speech.The Recall and Precision of the model reached 60% and 65%.(2)The speech corpus for medical environment was constructed.The spontaneous speech recognition system based on FP and HMM-GMM model was realized.The introduction of the FP detection model made the average CER% decreased by 1.94 and 2.37 for different test sets A and B.(3)The keywords extraction algorithm based on TF-IDF was studied.A keywords extraction algorithmwas proposed.which is based on the special structure and content of the medical record text.Experiments show that the Recall and Precisionof the algorithmboth can reach more than 60%.On this basis,the automatic matching of medical records based on the cosine similarity of medical record text was realized.(4)Designed and implemented the EMR system based on spontaneous speech recognition and keywords automatic extraction.The system achieved the oral medical record speech recognition,keywordsautomatic extraction,similar medical records automatically matching,automatic segmentation and FP clipping for long timespeech,multi process decoding and automatic annotation of punctuation Tests showed that the system has a good performance in daily use,the average CER% of spontaneousspeech recognition reached 85.09%.