A semi-automated DTI-based approach to evaluate structural characteristics and extension of cerebral gliomas (poster No C-2926)

Purpose: Diffusion Tensor Imaging (DTI) can identify peritumoral white-matter abnormalities in cerebral gliomas not apparent on conventional MRI, that can be referred to infiltration regions surrounding the tumor core. Our aim was to characterize pathological and healthy tissue in DTI datasets by 3D statistical texture analysis, with the purpose of developing a semi-automated detection technique (CAD) of cerebral tumors. Methods and materials: Fifteen patients with gliomas (9 low-grade, 6 high-grade) were selected. 3T MRDTI consisted of a single-shot EPI sequence (b=1000 s/mm2, 32 gradient directions). Diffusion maps were obtained (anisotropy maps: FA and q; isotropy maps: MD and p). Manual segmentation of pathological areas was performed on each map; 3D texture analysis was applied to these ROIs and to the contralateral healthy tissue, in order to identify discriminating features based on cooccurrence and “run length” matrices. Ninety features were calculated, with a sliding-window approach; the most representative ones were selected by the Fisher filter, and Principal Component Analysis was applied, followed by Neural Network training. Results: Six patients were employed for training, nine for testing. Sensitivity, specificity and ROC curves were calculated, giving satisfactory results (95% sensitivity at 88% specificity, ROC AUC 0.89). Test images were automatically segmented by tissue classification; manual and automatic segmentations were compared by the Jaccard coefficient, and were in good accord. Mapping of Principal Components was used to characterize the tumoral structure. Conclusion: This semi-automated approach looks promising for preoperative assessment of structural heterogeneity and extension of cerebral gliomas and for evaluating response to chemotherapy.