Research On Cervical Cancer Detection And Classification System Based On Diffuse Reflectance Spectroscopy

Cervical cancer is increasing and rejuvenating year by year,which seriously threatens women's life and health.Recently,there lacks a high efficient non-invasive detection method in clinic.In this paper,a set of cervical cancer detection and classification system based on diffuse reflectance spectroscopy has been proposed,which is convenient,portable and real-time.The neural network classification model has been built and upgraded to detect cervical cancer according to the several spectral features.The main work of this paper is as follows:1.The hardware platform for the inspection and classification system is designed,including the spectral acquisition subsystem and image acquisition subsystem.The spectral acquisition subsystem consists of a spectrometer,a light source,and a dual fiber probe for spectral data acquisition.The image acquisition subsystem comprises a CCD camera,a focus len and an endoscopic probe for providing spatial position information of the lesion of the cervical tissue.2.The software system based on commercial software Labwindows is developed,including acquisition module,processing module and auxiliary module.The acquisition module is used to collect and save spectral and image information.The processing module is utilized for spectral feature calculation,neural network model construction,and result prediction.Moreover,the auxiliary module is programmed to query database information and print reports.3.The calibration experiment and the animal verification experiment were conducted.The calibration of the tissue optical parameters was completed by a calibration experiment.The tumor characteristics of normal tissues and tumor tissues were verified by nude mice tumor experiments.The neural network model was constructed to distinguish between the two tissues.4.The clinical volunteer trial was designed.A method for detecting proper spectra is proposed.Then,the spectral features are extracted.The neural network and support vector machine models are established by using one vs.one and one vs.more classification strategies.Therefore,the model classification effects are compared.The research results show that the system developed in this study can be used for cervical cancer detection and classification,and the system has a potential for the clinical application.