PT - JOURNAL ARTICLE AU - Ko, Chi-Lun AU - Yen, Ruoh-Fang AU - Cheng, Mei-Fang AU - Wu, Yen-Wen AU - Tzen, Kai-Yuan TI - Combination of baseline 18F-FDG intra-tumor uptake heterogeneity indices derived from genetic algorithm predicts therapy response and survival in esophageal cancer DP - 2014 May 01 TA - Journal of Nuclear Medicine PG - 73--73 VI - 55 IP - supplement 1 4099 - http://jnm.snmjournals.org/content/55/supplement_1/73.short 4100 - http://jnm.snmjournals.org/content/55/supplement_1/73.full SO - J Nucl Med2014 May 01; 55 AB - 73 Objectives A good predictive marker is crucial for the better selection of patients with esophageal squamous cell carcinoma (ESCC), who may benefit from multimodality therapy. The present study utilized genetic algorithm (GA) to identify the predictive combination of PET derived heterogeneity indices and assessed its robustness in ESCC. Methods A total of 83 consecutive patients with locally advanced ESCC treated by neo-adjuvant chemoradiotherapy (CRT) and esophagectomy were included. Patients were randomly divided into the parameterization group (PG, n=42) and the validation group (VG, n=41). Global indices (such as SUVmax and metabolic tumor volume) and locoregional indices (those from co-occurrence, difference, and size-zone matrices) were extracted from baseline PET images. The best parameter combination was determined in PG by the customized GA, which optimized the prediction of pathological residual tumor after CRT. The combination was then validated in VG. The prognostic values of the combination regarding disease free survival (DFS) and overall survival (OS) were evaluated using Kaplan-Meier analysis. Results There were no significant differences in baseline characteristics between PG and VG. The absence of residual tumor after CRT predicted better DFS (P=0.006). Gray-level variability was the only index which predicted both residual tumor and survival in PG. However, it was neither predictive nor prognostic in VG. The GA identified the best combination (consisted of 1 local and 2 regional indices) in PG. The combination also predicted residual tumor in VG (accuracy=76%, P=0.02). In addition, it was prognostic regarding DFS and OS in both PG and VG (all P<0.05). Conclusions Each of the single indices from baseline FDG PET image was not powerful enough to be a predictive factor in this scenario. Nevertheless, GA was useful to identify the best combination of a few texture indices, which was a robust predictive factor of therapy response and survival in locally advanced ESCC.