Identification of Lung Tumors in Nude Mice Based on the LIBS With Histogram of Orientation Gradients and Support Vector Machine

Abstract

Early-stage detection of lung tumors helps to reduce patient mortality rates. In this work, we propose a method for diagnosing lung tumors in nude mice through combining laser-induced breakdown spectroscopy (LIBS) with the Histogram of Orientation Gradients (HOG) and Support Vector Machine (SVM). Firstly, the elemental spectral lines and elemental imaging maps for lung tissue are respectively obtained by the LIBS system. Secondly, the HOG is used to obtain the gradient direction relationship of multi-dimensional spectral intensity from LIBS images. The optimal spectral features based on HOG for different biological tissue can be extracted. And then, the SVM model is adopted to determine lung tumors. The results show that, compared to classification models based on SVM with full-spectrum emission intensity and SVM with Principal Component Analysis (PCA), the identification accuracy of lung tumors from the nude mice by using the HOG-SVM can be improved by 10.66% and 4.66%, the sensitivity can be improved by 12% and 4%, and the specificity can be improved by 8% and 6%, respectively. In addition, HOG-SVM is also used to differentiate inflammatory lung tissue from normal lung tissue in nude mice, and achieves the ideal classification result. This work shows that the LIBS technique combined with HOG-SVM provides a complementary method for the rapid detection of lung tumors, contributing to the successful treatment of patients.