TY - JOUR T1 - <strong>Radiomics technique helps to reduce the false positive rate of <sup>18</sup>F-FDG PET/CT diagnosis in suspicious lung cancer - importance of integrating clinical experience</strong> JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 55 LP - 55 VL - 60 IS - supplement 1 AU - Fei Kang AU - Wei Mu AU - Jie Gong AU - Weidong Yang AU - Wei Qin AU - Jing Wang Y1 - 2019/05/01 UR - http://jnm.snmjournals.org/content/60/supplement_1/55.abstract N2 - 55Purpose: Due to infective lesions, high false positive rate (FPR) of 18F-FDG PET/CT in lung cancer diagnosis is a severe challenge for making accurate clinical decision. Herein, based on novel radiomics methodology, we aim to establish a hybrid nomogram combining 18F-FDG PET/CT multimodality data and clinical judgement to reduce the FPR in the lung cancer differentiation of PET/CT, and compare its performance with the nomograms without clinical diagnosis integration. Methods: From 3947 screened lung-lesion patients who received 18F-FDG PET/CT scan from February 2007 to March 2017, 157 malignant and 111 benign patients were ultimately enrolled and randomly divided into training and test cohort with approximately equal sample size. Manual diagnosis was retrospectively recorded, and the reconstructed data of CT, thin-section CT (TSCT) and PET were stored. Lesion delineate and feature extraction were performed in training cohort following the radiomics methodology to establish and validate the nomograms of CT, TSCT, PET, PET/CT and PET/CT + manual diagnosis, respectively. The performance of these nomogram was assessed with respect to their calibration, key discrimination index, and clinical benefit in test cohort. Results: The overall FPR of manual diagnosis was 30.63%. Among the established nomograms, the C-index and YI of the PET/CT + manual diagnosis hybrid nomogram was the highest, and FPR the lowest in both of the training and test cohort (training and test cohort: C-index = 0.982 and 0.924, YI = 85.78% and 75.53%, FPR = 5.36% and 9.09%). In training and test cohort, by combining manual diagnosis, hybrid nomogram respectively corrected 78.57% (11/14) and 37.5% (3/8) FPR cases produced by the PET/CT RS. The net benefit of hybrid nomogram of PET/CT + manual diagnosis was the highest when the threshold probability is &lt; 85% (especially in the range of 60%-85%). Conclusions: Compared with manual diagnosis, the PET/CT + manual diagnosis hybrid nomogram is of superior differential accuracy of lung lesion. By adding manual diagnosis information, the hybrid nomogram could especially achieve lower FPR and higher clinical benefit than the radiomics nomograms solely based on imaging data. Key words: radiomics, 18F-FDG PET/CT, false positive rate, manual diagnosis, lung lesion differentiation ER -