Construction Of A Cancer-Related Fatigue Classification Model Based On Speech In Breast Cancer Patients

Objective:To investigate the overall status of cancer-induced fatigue in breast cancer patients and to understand their level of cancer-induced fatigue;The speech of breast cancer patients was collected to establish a data set for the classification of cancer-related fatigue in breast cancer patients.Radial basis function(RBF)support vector machine(SVM)was used to construct a classification model of carcinogenic fatigue in breast cancer patients and its performance was evaluated.Methods:1.The convenience sampling method was used to investigate the patients with breast cancer in the department of breast surgery of a third-grade hospital in Changchun.Research tools is the general information questionnaire and the Piper fatigue revised scale,application Epidata 3.1 and SPSS 21.0 to manage and analyze survey data,breast cancer patients in general information and revised Piper fatigue scale score using descriptive analysis,the influence factors of breast cancer patients with cancer-related fatigue analysis using single factor analysis method.2.Voice signals were collected from breast cancer patients in the department of Breast Surgery of a third-grade hospital in Changchun by semi-structured interview.The interview content included the current treatment received by the patients,their feelings of hospitalization,adverse reactions brought by treatment,emotional state,fatigue symptom experience and other aspects.Used Adobe Audition CC 2019 audio processing software for noise processing,editing and segmenting storage.Preweighting technology,framing technology,windowing technology and endpoint prediction technology were used to preprocess the voice signal,and then the acoustic characteristic parameters of the voice signal were extracted by openSMILE according to the extended the extended Geneva Minimalistic Acoustic Parameter Set.3.Used [0,1] interval normalization processing method for classification of breast cancer patients with cancer-related fatigue data set for data normalization processing,using LibSVM by 10 times 10 fold cross-validation method training and tuning based on radial basis kernel function of breast cancer patients with cancer-related fatigue model of support vector machine(SVM)classification,reuse of predict model set of test data to evaluate the performance of the model.Results:1.This study investigated a total of 110 patients with breast cancer,recycled effective questionnaire 101 in November 2018 to June 2019.101 breast cancer patients’ average age was 49.89 ± 9.57 years,80% of patients undergoing or received chemotherapy,an average of 3.50 ± 2.85 cycles of chemotherapy,60.40% incidence of cancer-related fatigue.101 patients with breast cancer-related fatigue median score of 3.27,cancer-related fatigue scores the highest in four dimensions is cognitive/emotional dimension.The 61 breast cancer patients with symptoms of cancer-related fatigue had an average score of 4.20 ± 1.24 for cancer-related fatigue.The results of univariate analysis showed that only age was the influencing factor of cancer-related fatigue,and the difference of scores of cancer-related fatigue in patients of different ages was statistically significant(P<0.05),and the degree of cancer-related fatigue in patients of younger age was more serious.2.This study to complete cancer-related fatigue status quo survey of 101 breast cancer patients who were voice signal acquisition,recording lasted about 15 ~ 60 minutes,after noise processing and editing,received no noise,complete sentence patterns and pronunciation length 1130 audio segments within 5 s,there was fatigue symptoms of speech segment is 542,the voice section of 588 patients with fatigue.According to the characteristics of acoustic characteristic parameters,all speech segments were selectively preprocessed,and the characteristic parameters of 88-uyghur tone were extracted from 1130 speech segments according to the expanded Geneva minimum acoustic parameter set,and the data set was constructed.3.In this study,the training set was used to obtain the classification model of carcinogenicity fatigue of breast cancer patients based on radial basis kernel function.The average accuracy of the training set was 84.99%,and the average number of correctly classified samples in the test set was about 96,with the average accuracy of 84.66%.After parameter selection and tuning,the average accuracy of 10-fold cross validation on the training set was 98.17%,the average number of correctly classified samples in the test set was 94,and the average accuracy of the test set was 82.39%.Conclusions:1.The overall level of cancer-related fatigue in breast cancer patients was moderate,and the degree of cancer-related fatigue in younger breast cancer patients was more severe,suggesting that medical staff should pay more attention to this part of the population.2.The average accuracy of the SVM classification model for breast cancer patients based on RBF is over 80%,which proves the effectiveness and practicability of the acoustic characteristic parameters of speech signals in predicting the symptoms of oncogenic fatigue.