Research On Disease-assisted Diagnosis Technology Based On Multimodal Medical Signals And Machine Learning

The application of machine learning methods in the field of disease diagnosis has been a frontier field that academia and industry have paid attention to in recent years,which can effectively solve the current problems of high misdiagnosis rate and backward diagnosis methods.Aiming at the problems in chronic renal failure(CRF)screening and parotid gland tumor diagnosis,this paper used the Fourier Transform Infrared Spectroscopy(FT-IR)signal of human serum samples and the computed tomography(CT)image signal of parotid gland tumors as the experimental research objects,combined with machine learning algorithms to establish a CRF screening model and a parotid gland tumor multi-class model.The main content of this paper includes:1.This paper applies serum FT-IR to the screening of CRF for the first time.In the experiment,the serum FT-IR of 44 CRF patients and 54 individuals with normal renal function were analyzed and compared,and partial least squares(PLS)was used to reduce the dimensions of original spectral data.Finally,support vector machine(SVM),extreme learning machine(ELM)and learning vector quantization(LVQ)algorithms were used to build diagnostic models.Through experimental comparison,it is found that the GS-SVM model achieves the best classification performance.And the accuracy of the model is 96.97%,the sensitivity is 93.75%,and the specificity is 100%.The experimental results show that the CRF diagnostic model proposed in this paper has good reliability,which also demonstrate that serum FT-IR combined with machine learning algorithms has great potential in screening CRF.2.This paper proposes a three-classification model of parotid gland tumors based on CT image features and machine learning algorithms.In this paper,the texture feature extraction method and the method of convolutional neural network model to automatically extract image features are respectively used on the CT images of parotid gland tumors,and a variety of classification algorithms are selected to construct the diagnosis model.In the best diagnostic model,the average area under curve(AUC)of parotid benign tumors(BTs)and malignant tumors(MTs)reached 0.8821,the average AUC value of pleomorphic adenomas(PAs)and Warthin tumors was 0.8132,the average AUC value of PAs and MTs was 0.9048,and the average AUC value of Warthin tumors and MTs was 0.8692.This shows that the auxiliary diagnosis model of parotid gland tumor proposed in this paper has good stability.The study in this paper shows that serum FT-IR combined with the PLS-SVM model can accurately distinguish CRF patients from those with normal renal function,which provides an effective,non-invasive and low-cost new idea for CRF screening.At the same time,it also indicates that the tri-classification auxiliary diagnosis model of parotid gland tumor established by the CT image features and machine learning algorithm can help doctors make a non-invasive preliminary diagnosis before surgery,thus reducing the workload of doctors to read the images,and providing a certain basis for clinical decisionmaking.