SU‐E‐J‐203: Texture Analysis of 3D and 4D PET/CT Images of Lung Cancer

Purpose: Texture/feature analysis of PET/CT images has great potential for predicting/monitoring tumor/normal tissue response to therapy, quantifying tumor spatiotemporal heterogeneity and radiosensitivity, and as indicators for adaptive therapy schemes. The objective of this study was to define a baseline and determine candidate texture/feature parameters. To this end, we performed image analysis on three dimensional (3D) and four dimensional (4D) PET/CT images of non‐small cell lung cancer patients. The ultimate goal is the development of radiomics tools for application in radiotherapy. Methods: All 4D‐PET data were reconstructed in 10 bins using the corresponding phases in 4D‐CT for attenuation correction. Co‐occurrence matrices of the tumor volumes were generated based on 3D and 4D PET images for 8 lung cancer patients. The texture parameters that were extracted from the original images or the co‐occurrence matrices included: entropy, local homogeneity, contrast, energy, V10‐V90, V80, V70, V40, I10‐I90, I30, Coefficient Variation, SD, Mean Intensity, Maximum Intensity, Minimum Intensity, and number of voxels. Texture comparisons between 3D and 4D PET and CT images and between various phases of 4D scans were performed. Results: A weak dependency of all the tested texture parameters on respiration phase was found for the 4D‐PET data for all patients, while large differences in some textures were found between the 3D‐PET or CT and one of the phases of the corresponding 4D data set, with larger percentage differences for the CT images. The differences for PET varied from 25% to 900%. Conclusion: Motion due to breathing altered statistical information obtained from intensity distributions from PET images, consequently 4D‐PET features were significantly different than 3D‐PET features. Texture parameters show minimal dependency on phase. PET images from one of the phases of a 4D‐PET with its corresponding 4D‐CT for attenuation correction are useful for texture analysis.