Correlation of ADC Texture With Treatment Response for Chemoradiation Therapy of Pancreatic Cancer International Journal of Radiation Oncology, Biology, Physics ORAL SCIENTIFIC SESSION| VOLUME 96, ISSUE 2, SUPPLEMENT , S70, OCTOBER 01, 2016
Date
10/01/2016Abstract
Purpose/Objective(s)
Imaging texture analysis has been shown to reveal tumor heterogeneity, which extends beyond what is visible to the human eye. The purpose of this study is to perform a texture analysis on apparent diffusion coefficient (ADC) maps obtained from diffusion weight MRI of pancreatic cancer and to correlate the texture with treatment response.
Materials/Methods
The MRI and pathological response data collected from 18 patients with resectable or borderline resectable pancreatic head cancer were analyzed. This set of patients all received neoadjuvant chemoradiation (nCR) before tumor resection. Pathological response to nCR was graded from the surgical specimen as G0, G1, G2, and G3 for complete, moderate, minimal, and poor response, respectively. The patient pool includes 4 patients for G1, 9 patients for G2, and 8 patients for G3 responses. Of the 18 patients, 17 have pre-nCR MRIs and 11 have post-nCR MRIs. The tumors were contoured based on multi-parameter MRI, including T1 Arterial, T1 Venous, and ADC images, on pre- and post-nCR MRIs. A statistical analysis was performed to identify regions of interest (ROI) and spatial trends between pre- and post-nCR MRIs. To detect statistical changes within a ROI, a t-test was used to compare the ADC spectrum of smaller ROI samples (ROIs) within the tumor to the overall distribution of ADC of the tumor ROI (ROIt). For each voxel, the ROIs is defined as N by N by N/3 region around the voxel. The P value of the t-test is mapped out to define the statistically similar regions, along with the outlier regions with higher or lower ADC mean than the ROIt mean. The statistically similar region is defined to be voxels within (1-P) < 0.0001; this translates to the distribution within the ROIs having a very similar mean and standard deviation as the ROIt distribution. For an evenly mixed combination of 2 means, the t-test will fail to isolate outliers due to the lack of variation between different ROIs. Additional textures were calculated in the identified ROIs, first order features to show the change in combination of cellularity and fibrosis, and GLCM features to show the fine variations of the texture.
Results
The ROIs identified with a statically similar manner between pre- and post-nCR ADC are different between the poor (G3) and moderate to good response (G1 and G2) cases. The mean percent of the total volume for G3 is 65 +/- 13% (pre-nCR) and 51 +/- 19% (post-nCR). For G2, 30 +/- 7% (pre) and 38 +/- 8% (post); and for G1, 39 +/- 9% (pre) and 38 +/- 8% (post).
Conclusion
Spatially, the ADC distribution for a G3 response tumor does not vary greatly in contrast to G1 or G2 response tumors which demonstrate greater inhomogeneity in both pre and post-nCR. These data reflect the texture of ADC maps acquired prior and post nCR could be used to assess tumor response and to improve the selection of patients for possible therapeutic intensification.
Author List
D. Schott W.A. Hall T. Gilat Schmidt E. Dalah K. Oshima E.S. Paulson P.M. Knechtges B.A. Erickson A. LiAuthor
Beth A. Erickson MD Professor in the Radiation Oncology department at Medical College of WisconsinView Online
