A Pilot Study of Radiomics Models Combining Multi-Probe and Multi-Modality Images of (68)Ga-NOTA-PRGD2 and (18)F-FDG PET/CT for Differentiating Benign and Malignant Pulmonary Space-Occupying Lesions

BACKGROUND: This is a pilot study of radiomics based on (68)Ga-NOTA-PRGD2 [NOTA-PEG4-E[c(RGDfK)]2)] and (18)F-FDG PET/CT to (i) evaluate the diagnostic efficacy of radiomics features of (68)Ga-NOTA-PRGD2 PET in the differential diagnosis of benign and malignant pulmonary space-occupying lesions and (ii) compare the diagnostic efficacy of multi-modality and multi-probe images. METHODS: We utilized a dataset of 48 patients who participated in (68)Ga-NOTA-PRGD2 PET/CT and (18)F-FDG PET/CT clinical trials to extract image features and evaluate their diagnostic efficacy in the differentiation of benign and malignant lesions by the Mann-Whitney U test. After feature selection with sequential forward selection, random forest models were developed with tenfold cross-validation. The diagnostic performance of models based on different image features was visualized by receiver operating characteristic (ROC) curves and compared by permutation tests. RESULTS: Fourteen of the (68)Ga-NOTA-PRGD2 PET features between benign and malignant pulmonary space-occupying lesions had significant differences (P<0.05, Mann-Whitney U test). Eighteen of the (68)Ga-NOTA-PRGD2 PET features demonstrated higher AUC values than all CT features in the differential diagnosis of pulmonary lesions. The AUC value (0.908) ​​of the three-modal feature model was significantly higher (P<0.05, permutation test) than those of the single- and dual-modal models. CONCLUSION: (68)Ga-NOTA-PRGD2 PET features have better diagnostic capacity than CT features for pulmonary space-occupying lesions. The combination of multi-modality and multi-probe images can improve the diagnostic efficiency of models. Our preliminary clinical hypothesis of using radiomics based on (68)Ga-NOTA-PRGD2 PET images and multimodal images as a diagnostic tool warrants further validation in a larger multicenter sample size.