Detection of the gene mutation of epidermal growth factor receptor in lung adenocarcinoma by radiomic features from a small amount of PET data.

Abstract

The purpose of this work was to identify the potential mutation of epidermal growth factor receptor in nonsmall cell adenocarcinoma by noninvasive method, and to explore whether the same or better effect can be achieved using a small amount of single-mode PET image data. A total of 115 patients were recruited and the results of their 18F-FDG PET images and gene detection after resection were obtained; 117 original radiation features and 744 wavelet transform features were extracted from PET images. Several methods were used to reduce the dimension of the data, and four classifier models were established to classify it. The above process was repeated to reduce the total amount of data and the area under the receiver operating characteristic curve (AUC) value that changed with the reduction of the data and the stability of the results was recorded. The classifier with the best comprehensive performance under this dataset was logistic regression, whose AUC value is 0.843. And similar results can be obtained from only 30 cases of data. A similar or better result could be achieved using a small number of single-mode PET images. In addition, significant results could be obtained using only the PET images of 30 patients.