Research On Non-invasive Measurement Of Optical Parameters Of The Tissues By Support Vector Machine

The interaction between the laser and biological tissues is represented by a series of optical response, such as absorption, scattering, etc., so the outgoing light of tissues surface carries the physiological and pathological states inside the tissue, these pathological information is often reflected in the tissue optical characteristic parameters, tissue optical parameters obtained from the diffuse reflection light distribution of tissue surface is the most commonly used method to provide convenient conditions for the practical application of medical diagnosis and treatment.The common method for determining the optical parameters is Diffusion Approximation Theory and Neural Network, so they have had a better application in biological tissue. Support Vector Machine(SVM) is used to study the optical parameters of tissues. Firstly, Monte Carlo(MC) is applied to obtain the data; Then these data are used to establish the relationship model between the diffuse reflection light distribution and the optical parameters; Lastly, the established model is used to predict the optical parameters of the tissue.In the research of monolayer tissue optical parameters, we examine its impact on the accuracy of the model prediction from the changes in the number of data samples, pretreatment methods and test set and the training set contains relations firstly. Resulting that:1.the test set is not required to include within the scope of the training set, namely the SVM model once established, the corresponding regression function get by it; 2. Choose samples close to the test set of data as the training set, the model prediction accuracy is higher; 3. With the reduction of the number of samples of the training set, the modeling time ever shortening and increasing the prediction error, but also higher prediction accuracy in the case of a small sample; 4. After data preprocessing, SVM modeling time is significantly shorter;5. Sample sets were obtained by using MC data and CCD video device to established SVM model. Finally the optical parameters of the tissue simulating liquid was predicted from the light distribution of the diffuse reflectance of the surface of the tissue simulating liquid by using the model established before. The average relative error of t? were 14.58% and 9.97%; The average relative error of eff? were 15.98% and 10.66%, respectively. The results show that model established by using experimental data can get a higher prediction accuracy of the optical parameters, and the SVM method has a good real-time to the prediction of optical parameters.For the study of the optical parameters of the bi-layer organization, due to there is not unique relationship between the diffuse reflected light and the optical parameters. the first to study the variation of the diffuse reflected light intensity tissue optical parameters under different conditions, namely fixed layer of tissue optical parameters, light distributed scattering coefficient changes with another layer of three trends: monotonic rise, non-monotonic and monotonic decline; decreases monotonically with another layer of change in the absorption coefficient of a trend: monotonic decline. According to these laws the use of SVM diffuse reflected light distribution and optical parameters to establish a better relationship model, using the model prediction of a double tissue optical parameters, the results show that the average relative error of the optical parameters are: For the uppera1? : 2.3 %, s1? : 3.15%; lower a2? : 5.8%,s2? : 4.43%; For the thickness of the upper layerl : 5.2%. It can be seen, the use of SVM to establish the predictive accuracy of the model is ideal.The innovation is that the relationship model between organizational light distribution and optical parameters is built by SVM. We give enough discussion for single-layer tissue and preliminary discussion for double-layer tissue; These studies is helpful for further discussion. In practice, the optical parameters can be get though this model directly, thus it satisfied the requirements of real-time.