Room: Exhibit Hall | Forum 5
Purpose: Assessment of spatial distribution of tumor microenvironment is critical for the evaluation of treatment effectiveness or tumor recurrence. This study explores the feasibility of using radiomic features to spatially differentiate internal tumor environment on CT images.
Methods: Metabolically active and inactive regions were separately contoured on CT data using PET/CT fused images for four patients. The PET-cold regions were identified using CT data alone. A total of 1419 radiomic features from each contour were computed using IBEX software on CT image data. Mean, standard deviation, percentage change and contrast-to-noise ratio (CNR) of all the features were calculated, and then compared between demarcated metabolically active and inactive regions on the tumor using CT images. A threshold of 98% was used to estimate most sensitive/ useful features from the list.
Results: 28 features were identified based on CNR (98% cut-off) computation, which can differentiate the tumor into metabolically active and inactive regions on the CT images. The features are Gray-Level Cooccurrence Matrix (Entropy (10-7, 5-7, 9-7), SumEntropy (10-7, 5-7), SumVariance (5-7, 10-7, 9-7), AutoCorrelation (5-1, 2-1, 9-1, 12-1, 6-1, 7-1, 10-1), SumAverage (2-1, 9-1, 12-1, 7-1, 6-1)), Intensity direct/ intensity histogram (Percentile (55, 50), Global median, Quantile (0.5)). The calculated CNR values for the features above threshold were ranging from 1.98- 4.83HU. The mean percentage changes calculated between the two regions was 30% (10-7Entropy), 25% (5-7Entropy), 24% (9-7Entropy), 23% (10-7SumEntropy), and 26% (5-7SumEntropy).
Conclusion: The extracted radiomic features for the spatial differentiation of tumor compartments were robust and the results demonstrated the potentiality of using radiomic features for the spatial separation of lung tumors. We will further expand the database by increasing sample size and implement radiomics immensely in our future work.