Study And Application Of Chemometrics New Methods In Data Analysis And Diagnosis Of Clinical Diseases

Metabolite analysis and medical imaging are two commonly used medical examination methods,which can produce a large amount of clinical metabolomics data and medical images.How to extract disease-related features from these complex data and distinguish the disease and healthy samples is a hot issue in clinical disease diagnosis.In this thesis,by studying the application of chemometrics methods in metabolomics data analysis and medical image analysis,the author proposes several auxiliary clinical disease diagnosis strategies.This thesis mainly includes the following aspects:Inherited metabolic diseases（IMDs）can cause corresponding biochemical metabolic disorder of newborns and even sudden infant death.Timely detection and diagnosis of IMDs are of great significance for improving survival of newborns.Here we propose a strategy for simultaneously detecting six types of IMDs based on random forest（RF）algorithm.Clinical urine samples from IMD and healthy patients are analyzed using GC-MS for acquiring metabolomics data.Then,the RF model is established as a multi-classification tool for the GC-MS data.Compared with the models built by artificial neural network（ANN）and support vector machine（SVM）,the results indicate that the RF model has superior performance of high specificity,sensitivity,precision,accuracy,and matthews correlation coefficients on identifying all six types of IMDs and normal samples.The proposed strategy can afford a useful method for reliable and effective identification of multiple IMDs in clinical diagnosis（In Chapter 2）.There are mainly two tasks in metabolomics data analysis,disease classification and biomarker discovery.Here we propose a strategy of assembling discrete particle swarm optimization（DPSO）into stacked autoencoder（SAE）to form a framework called DPSO-SAE for the classification of two types of IMDs and healthy samples based on the GC-MS urine metabolomics data.Superior performance is obtained by DPSO-SAE with high accuracy,good generalization ability on classification and robustness in variable selection.Six potential biomarkers are proofed for two types of IMDs.The proposed strategy of DPSO-SAE,therefore,may provide a valuable modeling algorithm for disease diagnosis based on metabolomics data analysis（In Chapter 3）.Knee osteoarthritis（OA）is one of the common joint diseases caused by the corrosion of interarticular cartilage,making the kneecap contact and friction.It is one of the most common diseases in the elderly population.We propose a method based on faster R-CNN for automatic detection of the knee joint and KL（Kellgren-Lawrence）grading of knee OA.Faster R-CNN is a two-step deep object detection convolutional neural network,which uses RPN（region proposal network）to replace the time-consuming selective search method.RPN has fewer regional proposals and higher quality.We train and test the constructed model on the OAI（Osteoarthritis Initiative）public dataset.Compared with some other reported knee detection methods,we obtain better accuracy,m AP,recall and F₁score.Therefore,this method can be used as an auxiliary means for clinicians to diagnose the disease initially（In Chapter 4）.Building on our previous work,we propose a novel modeling strategy based on YOLO v3（You Only Look Once version 3）for simultaneous localization of the knee joint and KL classification of knee OA.YOLOv3 is an advanced one-step object detection convolutional neural network algorithm.Compared with faster R-CNN,YOLOv3 has a faster detection speed and can elegantly integrate the detection of knee joint and the grading of OA severity.In addition,in this work,we added a part of self-collected clinical data sets（knee X-ray images from the Third Xiangya Hospital）to participate in the training and validation of the model based on the public data sets.The results indicate that the accuracy,recall,F₁score and diagnostic accuracy of knee OA are improved.It provides a convenient and efficient image analysis for serving daily clinical diagnosis（In Chapter 5）.