Room: Exhibit Hall | Forum 5
Purpose: To quantify the difference of Â¹â?¸F-FDG PET radiomic features between non-small cell lung carcinoma (NSCLC) subtypes with adenocarcinoma (ADC) and squamous cell carcinoma (SqCC).
Methods: Diagnostic Â¹â?¸F-FDG PET scans of 97 patients (median age,63; range,36-85) with NSCLC (61 ADC and 36 SqCC) were retrospectively analyzed, and pathological type was determined by histopathologic examination after biopsy and surgery. Combined the SUV threshold and manual segmentation method, the gross target volume (GTV) was delineated. Then, 49 metabolic features and 59 image features were extracted from the GTV using the radiomic method. Comparing differences of radiomic features between ADC and SqCC, the ability to identify tumor subtypes were evaluated by the receiver operating characteristic curve (ROC) and the area under curve (AUC) value greater than 0.7 was considered predictive.
Results: A total of 108 features were extracted. The number of radiomic features that differed statistically significant between ADC and SqCC was 20, including 5 metabolic features and 15 image features. The AUC value of features with statistically significant differences were all greater than 0.7, and these features have strong correlation to the pathological type (AUC range:0.704-0.777, all p-value<0.05). Features with higher AUC value showed higher Youden value (Short run emphasis, AUC=0.765, Youden index=0.522; Contrast, AUC =0.758, Youden index=0.527; Intensity, AUC=0.773, Youden index=0.55). Meanwhile, the average AUC value of image features (0.760Â±0.02) was higher than metabolic features (0.747Â±0.03).
Conclusion: Some of the radiomic features has significant difference between ADC and SqCC. Image features performed better accuracy of pathological typing than metabolic parameters. Thus, radiomic features derived from Â¹â?¸F-FDG PET images can be served as subtype biomarkers.