Comparison of PET conventional parameters and texture parameters in prediction of KRAS gene mutation in colorectal cancer / 中华核医学与分子影像杂志

ABSTRACT

Objective To evaluated the value of PET conventional parameters and texture parameters in prediction of the Kirsten rat sarcoma viral oncogene (KRAS) gene expression status in colorectal cancer (CRC) by analyzing the relationship between those parameters and KRAS gene status. Methods From December 2012 to January 2017, 18 F-fluorodeoxyglucose (FDG) PET/ CT data and KRAS gene status of 58 CRC patients (40 males, 18 females, average age 56.31 years) before anti-tumor therapies were collected. The relation-ship between PET parameters and KRAS gene expression was analyzed. Receiver operating characteristic (ROC) curve analysis was used to evaluate the values of PET conventional parameters and texture parame-ters for predicting the KRAS gene status. Spearman rank correlation and Mann-Whitney u test were used to analyze the data. Results Of 58 CRC patients, 19(32.8％) had KRAS mutation, while 39(67.2％) were with wild type KRAS. Among the 46 PET image parameters extracted by Chang-Gung image texture analysis (CGITA), 14 PET image parameters were selected by Spearman rank correlation (all ︳rs︳>0.8), inclu-ding 12 texture parameters and 2 conventional parameters (maximum standardized uptake value (SUVmax ) and total lesion glycolysis (TLG)). Six PET image parameters (4 texture parameters and 2 conventional pa-rameters) were found to be different between KRAS gene mutant group and wild group (u values from -4.481 to -2.046, all P < 0. 05). Among the 4 texture parameters, standardized uptake value ( SUV) kurtosis (SUVkur ) had the best prediction value, while SUVmax was the better one for prediction in the 2 conventional parameters. When 4.27 was selected as the cut-off value for SUVkur , the Youden index was up to the maxi-mum as 0. 35 and it yielded 0.667 on the area under curve (AUC) (95％ CI 0.517-0.816, P = 0.041) with the sensitivity of 15/ 19 and specificity of 56.4％(22/ 39), respectively. When 16.6 was selected as the cut-off value of for SUVmax , the Youden index was up to the maximum as 0.64 and the AUC on predicting the KRAS mutant was 0.865 (95％ CI 0.770-0.960, P<0.001) with the sensitivity of 17/ 19 and specifici-ty of 74.4％(29/ 39), respectively. The efficacy of SUVmax for predicting KRAS mutation was significantly better than that of SUVkur(z= 3.258, P= 0.001). Conclusion PET texture parameters and conventional pa-rameters can be used to predict the KRAS gene status in CRC patients, and SUVmax may be the best parameter.