Molecular Imaging Visualization of Tumor Heterogeneity in Non-small Cell Lung Cancer

In this study, the investigators first selected the features with significant differences between genes mutant and wild type in the training set using the Mann-Whitney U test, obtaining a total of 53 features with p value < 0.05. Then, the least absolute shrinkage and selection operator (LASSO) algorithm was used to select the optimal predictive features among the 53 selected in the training set. The LASSO algorithm adds a L1 regularization term to a least square algorithm to avoid overfitting. A prediction model was established by logistic regression, and the radiomics signature score (rad-score) for each participant was calculated based on the selected discriminating radiomic features. The model performance was tested in the validation set. The receiver operating characteristic (ROC) curve and the area under the curve (AUC) were used to evaluate the model performance in the training and validation sets.

Genes mutation detection was performed on histological specimens obtained by surgical resection or puncture. Genes mutation detection was conducted by the examination of the whole-exome sequencing. The whole-exome sequencing is performed by the Novogene company (www.novogene.com). If any genes mutation in the whole-exome sequencing was detected, the tumor was considered genes mutant. Otherwise, the tumor was classified as no mutant of genes type.

The imaging acquisition protocol was established following the Image Biomarker Standardisation Initiative (IBSI) reporting guidelines. All the details of the procedures are provided in the electronic supplementary material. 18F-FDG PET/CT (Discovery MI, GE Healthcare, Milwaukee, WI, USA) examination was performed within 1 month before treatment. Briefly, 18F-FDG PET/CT images were acquired 60 ± 5 min after 18F-FDG injection in accordance with the European Association of Nuclear Medicine (EANM) guidelines, version 1.0 (version 2.0 from February 2015). A 3D LOR reconstruction algorithm was used to reconstruct the PET images. The attenuation correction of PET images was carried out with CT data, and the corrected PET images were fused with CT images.

The LIFEx package (version 4.00, http://www.lifexsoft.org) was used to extract the texture features of PET/CT images of lesions in the same VOI. The PET/CT image of the patient in the DICOM format was imported into the software. Two experienced PET/CT diagnostic physicians semi-automatically delineated the region of interest of the target lesion using a threshold of 40% of the maximum standardized uptake value (SUVmax). Then, the software program automatically calculates and extracts 47 PET radiomic features and 45 CT radiomic features.