Study On The Key Techniques Of Human Gastric Cancerous Tissue Detection Based On Terahertz Spectroscopy

As a burgeoning branch of vibrational spectroscopy,terahertz technology has recently drawn extensive attention in biomedical research area,especially in human cancer diagnosis,for its exclusive detection advantages of being fast,non-destructive and sensitive to biomedical molecules and water.Even though the terahertz research on medical diagnosis has received great achievements till now,further investigations are needed to study the mechanism and problems in practical applications.With the support of the National Natural Science Foundation,this dissertation takes human gastric cancer as research object and investigates the techniques of applying terahertz radiation to the human cancer detection.In this work,a comprehensive experimental study on terahertz response of human cancerous tissue and contrast mechanism is carried out.The identification of terahertz absorption spectra of cancerous tissue is primarily studied.To that end,two classification methods are proposed from different perspectives to improve the performance and interperetability of the terahertz technology in medical diagnosis.One is an optimized pattern recognition model which is based on characteristic extraction of principle components of terahertz spectra.The other is a discrimination method out of spectrum source analysis using spectral unmixing technqiues.The main contributions and innovations of this dissertation are summarized as follows:(1)With the guidance of the medical theory of human tissue carcinogenesis,a comprehensive experiment study,which investigates the effect of the pathological changes in cancerous tissue,including the alteration of tissue morphology,hydration state and biological component on cancer detection by terahertz time-domain spectroscopy,is initially conducted to explore the suitable sample preparation method and the source mechanism of tumor terahertz response.Meanwhile,the experimental techniques of terahertz detection on tissue are discussed for helping obtain valid terahertz measurement signal of tissue.(2)Targeted to the terahertz absorption spectra of tissue,a classification model based on the characteristic extraction of principle component of terahertz spectra is proposed,which can automatically classify and identify the pathological state of tissue by the optimized pattern recognition method.This model makes use of principle component analysis incorporating with t-test to reduce the dimension and extract features of the absorption spectra with supervision so that the extracted features have the specificity of highlighting terahertz signal contrast between two kinds of tissue.Besides,the classification algorithm adopted in the model considers the overfitting problem resulting from the small tissue sample size at present.The experimental results show that the presented classification model of terahertz absorption spectra of tissue can identify gastric cancer precisely with a small sample size and have strong generalization ability.(3)In order to further analyze the mechanism of the terahertz absorption spectra variation corresponding to the gastric tissue carcinogenesis and excavate the pathological information hiding in the tissue spectra like tumor markers or characteristic molecules,we investigate the resolution techniques of terahertz absorption spectra of mixtures.The terahertz absorption spectrum of a multicomponent mixture is likely to consist of overlapping absorption bands from individual components,which leads to the interference and submergence of identification information of components.To solve that problem,we adopt two kinds of blind source separation methods,i.e.,Nonnegative Matrix Factorization(NMF)and Hard Modeling Factor Analysis(HMFA),to identify the component spectra of unknown mixtures separately from the point of mathematic analysis and the physical structure decomposition of spectra.Moreover,we modify and improve the two algorithms to fix the problems arising in the application to the resolution of terahertz spectra.The nonuniqueness of NMF is improved by adding the constraint condition of smoothness of terahertz spectra.The rule for choosing distincitive peaks in HMFA is modified to make it get rid of the inference of artificial baseline over low frequencies.Constrained NMF is easy to realize and has no requirement of spectra but the number of components.Comparatively,the implementation of improved HMFA is more complicated and it demands that there is at least one distinctive peak in every component spectra.Nevertheless,this algorithm has clear physical indications.The two methods are validated with the simulated and experimental terahertz absorption spectra of mixtures of amino acids.Results indicate that the extracted component spectra by those methods all agree well with the true spectra.Improved NMF and HMFA methods achieve the overlapping terahertz spectra unmixing and provide useful tools for deep exploration of terahertz spectra of human tissue and other complicated mixtures.(4)For the sake of improving the interpretability of cancer detection by terahertz techniques,the link between terahertz absorption spectra of tissue and pathological mechanism is studied and an identification method for tissue types based on spectra resolution is put forward.This method makes attempt to distinguish cancer tissue by the way of spectral mechanism analysis.It firstly decomposes the terahertz absorption spectra of normal and cancerous tissue into the weighted sum of peak functions and extracts the characteristic peaks implying the occurrence of cancer.Those peaks may correspond to the intrinsic molecules or molecular modes whose quantity or structure change remarkably during the process of carcinogenesis.Finally it combines with probability density distribution and CART to decide the peak weight threshold and sets the classification rules according to the decomposed peak weight distributing discipline of the two kinds of tissue spectra.This method can not only discriminate the gastric tissue types but also resolve the hidden characteristic absorption peaks correlated to the tumor markers.The presented tissue classification method based on spectra resolution has the advantage of low computational complexity and serves as a new research route for terahertz detection of diseased tissue samples in large scale.The suggested approach is expected to help make medical diagnosis by the terahertz technology quick and simple like spectroscopy detection,as well as reliable and interpretable like pathological examination.In summary,amiming at the problems of the mechanism exploration and the application of terahertz technology to cancer detection,this thesis investigates the key techniques including experimental skills and analysis of signal sources,classification of tissues based on pattern recognition,resolution of overlapping terahertz spectra and identification of cancer tissue based on excavation of spectral sources.The achievements of this research provide technical support for cancer detection by the terahertz technology and prompt the application of terahertz techniques to practical medical diagnosis,which can effectively complement the current cancer diagnosis tools and aid the diagnosis.The tissue samples and their pathological analysis in this research are provided by the Second Affiliated Hospital of Zhejiang University School of Medicine.