Program Information
Quantitative Imaging Biomarker Reproducibility in Pancreatic Cancer Longitudinal CT Texture Analysis Study
S Klawikowski*, J Christian , D Schott , X A Li , Medical College of Wisconsin, Milwaukee, WI
Presentations
WE-F-205-10 (Wednesday, August 2, 2017) 1:45 PM - 3:45 PM Room: 205
Purpose: This work aims to assess the impact of delineated structure variation on CT texture analysis based on longitudinal CTs collected during CRT (chemoradiation therapy) for pancreatic cancer.
Methods: First and last daily CTs acquired during CRT for 16 pancreatic head cancer patients using in-room CT were analyzed. The pancreas head was manually contoured on each daily CT. MIM® software was used to duplicate and slightly modify each initial pancreas head contour, simulating the typically observed inter- and intra-observer variations on structure delineation. A total of 22 contour manipulations included translations by 1-3 voxels in various directions and shrinking/expanding in small voxel increments. Texture analysis was then performed using publically available software (IBEX). 1453 texture metrics including: grey level co-occurrence, run-length, histogram, neighborhood intensity difference, and geometrical shape features were calculated for each daily CT for each contour. Metrics with non-numeric values and co-occurrence metrics with length scales larger than 4 voxels were removed from further analysis leaving the total number of metrics analyzed at 990. A normalized standard deviation of the mean was calculated for each metric running over all 32 daily CT’s and 23 contours to quantify the relative individual texture sensitivity to slight changes in contour geometry.
Results: First order histogram metrics, shape metrics, and neighborhood intensity metrics were found least sensitive (normalized standard deviation of the mean) to small contour variation (≤0.221, ≤0.016, ≤0.003 σ/(√N*AVG(x))). Co-occurrence contrast and difference-entropy metrics were found to be most sensitive to small geometrical contour changes. (≤0.9416, ≤0.9416 σ/(√N*AVG(x))). These metrics are especially sensitive to contours expanding the pancreas head boundaries and thus sensitive to the delineation accuracy of the boundaries.
Conclusion: Impact of inter- and intra-observer contouring variations on CT texture analysis is relatively small on the first-order features, and, however, can be significant for higher order features.
Contact Email: