SU‐E‐J‐265: Feasibility Study of Texture Analysis for Prognosis of Local Tumor Recurrence Within 5‐Years for Pharyngeal‐Laryngeal Carcinoma Patients Received Radiotherapy Treatment

Abstract

Purpose:Pharyngeal and laryngeal carcinomas (PLC) are among the top leading cancers in Asian populations. Typically the tumor may recur and progress in a short period of time if radiotherapy fails to deliver a successful treatment. Here we used image texture features extracted from images of computed tomography (CT) planning and conducted a retrospective study to evaluate whether texture analysis is a feasible approach to predict local tumor recurrence for PLC patients received radiotherapy treatment.Methods:CT planning images of 100 patients with PLC treated by radiotherapy at our facility between 2001 and 2010 are collected. These patients were received two separate CT scans, before and mid‐course of the treatment delivery. Before the radiotherapy, a CT scanning was used for the first treatment planning. A total of 30 fractions were used in the treatment and patients were scanned with a second CT around the end of the fifteenth delivery for an adaptive treatment planning. Only patients who were treated with intensity modulated radiation therapy and RapidArc were selected. Treatment planning software of Eclipse was used. The changes of texture parameters between two CT acquisitions were computed to determine whether they were correlated to the local tumor recurrence. The following texture parameters were used in the preliminary assessment: mean, variance, standard deviation, skewness, kurtosis, energy, entropy, inverse difference moment, cluster shade, inertia, cluster prominence, gray‐level co‐occurrence matrix, and gray‐level run‐length matrix. The study was reviewed and approved by the committee of our institutional review board.Results:Our calculations suggested the following texture parameters were correlated with the local tumor recurrence: skewness, kurtosis, entropy, and inertia (p<0.0.05).Conclusion:The preliminary results were positive. However some works remain crucial to be completed, including addition of texture parameters for different image features, sensitivity of tumor segmentation variations, and effect of image filtering.