Purpose/Objective(s): The aim of this study was to determine whether texture analysis of FDG-PET/CT is a predictor of outcomes of definitive chemoradiotherapy (dCRT) for patients with stage II - III thoracic esophageal cancer.

Materials/Methods: We included patients who were treated between 2005 & 2013 and underwent FDG-PET/CT within 45 days before dCRT at our institution. Patients who had massive pneumonia caused by perforation of the tumor at FDG-PET/CT & patients in whom the metabolic tumor volume (MTV) of the primary tumor were less than 10 ml were excluded. 46 patients were enrolled in this study. We measured the primary tumor’s MTV by using a commercially available deformable registration algorithm and calculated the texture parameters in the MTV using a computer algorithm. Texture parameters that had strong correlation with TLG whole body & with each other texture parameters were excluded. Survival estimates were calculated using the Kaplan-Meier method from the first date of dCRT. Predictors were analyzed using Cox’s hazards model.

Results: The median follow-up period was 27.7 months & the 3-year overall survival (OS) rate, local control (LC) rate and progression-free survival (PFS) rate were 49.7%, 32.7% & 30.4%, respectively. In multivariate analysis, correlation (co-occurrence matrix) was an independent predictor for OS (p= 0.0067). The number of cycles of concurrent chemotherapy (1 cycle vs 2 cycles, p = 0.0061), inverse difference moment (normalized co-occurrence matrix, p= 0.0014) & Correlation (co-occurrence matrix, p= 0.035) were independent predictors for LC. The number of cycles of concurrent chemotherapy (1 cycle vs 2 cycles, p= 0.0074), correlation (co-occurrence matrix, p= 0.001) & code similarity (texture feature coding co-occurrence matrix, p= 0.034) were independent predictors for PFS.

Conclusion: Correlation was an independent predictor for OS, LC & PFS. Inverse difference moment was an independent predictor for LC, & code similarity was an independent predictor for PFS. It is possible that texture analysis will become an important method to predict outcomes of dCRT for patients with esophageal SCC.

Multivariate analysis
	OS		LC		PFS
Variables	HR (95% CI)	p value	HR (95% CI)	p value	HR (95% CI)	p value
Gender (male vs female)	3.00 (0.47-59.1)	0.274
Chemo 1 cycle vs 2 cycles			6.63 (1.78-23.6)	0.0061	6.27 (1.69-22.2)	0.0074
TLGWB*	2.45 (0.62-9.68)	0.199
Intensity variability* (Voxel-alignment)	1.44 (0.33-6.31)	0.632
High-intensity run emphasis* (Voxel-alignment)			0.76 (0.32-1.79)	0.519	0.92 (0.39-2.19)	0.852
Contrast* (Neighborhood intensity-difference)	2.67 (0.54-15.4)	0.235
Short-zone emphasis* (Intensity-size-zone)	0.79 (0.24-2.40)	0.683
Inverse difference moment* (Normalized co-occurrence)	2.33 (0.92-6.55)	0.074	3.78 (1.64-9.46)	0.0015	2.08 (0.82-5.39)	0.122
Correlation* (Co-occurrence)	4.25 (1.47-13.6)	0.0067	2.29 (1.06-5.10)	0.035	3.84 (1.72-8.90)	0.001
Code Similarity* (Texture Feature Coding co-occurrence)					0.33 (0.12-0.92)	0.034
note: * ≥ median vs < median