Detection of Lung Cancer with Breath Biomarkers Based on SVM Regression

According to feature extraction of high order cumulant, a new method of detecting lung cancer is proposed applying support vector machine model to recognize the mixed volatile organic compound (VOC) infrared spectrum, where the primary and secondary absorbed peaks are seriously overlapped. The number of spectrum channel of the original spectrum data is large; hence, the transmitted spectrogram is mapped to four-order cumulant space and detached from each firstly. In this simulation experiment, concentration of 19 VOCs was regressed by SVM and the result shows that the method performed well in identification. The average correct rate of component recognition is more than 95.5% when component concentration of VOC is not less than 1%. MSE and MAE were introduced to assess the performance of the method. Prediction adopting SVM and ANN is compared.