Automated spectroscopic tissue classification in colorectal surgery The number of extracted features (n=36) is too large to perform a statistically meaningful classification, as the extracted features could be redundant in the information they retain. Therefore, using combinations of all 36 features to build a classifier would result in a dimensionality problem and over‐fitting31. We identified the best distinctive features for classification of ureter/artery in an adipose surrounding by using binary logistic regression (SPSS Inc., Chicago, IL, USA) for both wavelength regions separately. This concerns a statistical technique that allows prediction of categorical dependent variables (i.e. the tissue type: ureter/artery versus adipose tissue) using a set of independent variables (i.e., the spectral features)32. Using the approach for tissue classification based on hyperspectral data reported earlier by Akbari et al.22, we used support vector machine (SVM) to classify ureter/artery within adipose surrounding. The SVM classifier was trained with a polynomial kernel function33. This classifier attempts to find an optimum line to separate the training data groups with a minimum risk31. To estimate classification performance and to prevent overly optimistic results34‐36, we implemented a cross–validation (CV) approach. With a goal to obtain the classification accuracy as a performance measure, the data set is divided into a training set (to train the classifier) and a test set (to validate the classifier). The CV approach uses leave‐one‐out (LOO) validation of the acquired ureter/artery and adipose hyperspectral data. This approach utilizes the same data set for both training and testing purposes and is very useful in cases of a relatively small data sample. In‐house developed classifiers (using MATLAB environment Version 7.7.0, MathWorks Inc., Natick, Massachussetts, USA) were used to estimate the classification performance by calculating sensitivity, specificity, positive predictive value, negative predictive value and accuracy for both wavelength regions (i.e. Si‐sensor and InGaAs‐sensor detection range). The methods used for data‐processing of the in vivo hyperspectral data in this study, have previously been applied for constructing a classification system for automated nerve differentiation within adipose surrounding in another in vivo study27. 113
proefschrift_Schols_SLV
To see the actual publication please follow the link above