Louis Gagnon

We present the University of California San Diego post-treatment glioblastoma (UCSD-PTGBM) annotated multimodal MRI dataset. The UCSD-PTGBM dataset includes 243 timepoints on 178 subjects with histopathologically-proven glioblastoma who were imaged with an advanced brain tumor protocol on 3 Tesla MRI scanners. Sequences include standard 3D imaging, as well as multishell diffusion (Restricted Spectrum Imaging, RSI) and perfusion imaging techniques (Arterial Spin Labelling, ASL and Dynamic Susceptibility Contrast, DSC), and neuroradiologist approved voxelwise tumor segmentations for both traditional segmentation masks and cellular tumor segmentations. The dataset also includes isocitrate dehydrogenase (IDH) mutation status and O6-methylguianine-DNA methyl-transferase (MGMT) promotor methylation status, as well as overall survival and progression free survival information for a subset of cases. We hope that researchers around the world will use these data to continue to improve analysis of post-operative MRI on glioblastoma patients, translate their findings into clinical practice and improve the management and outcome for these patients.

Background: Differentiating recurrent tumor from post-treatment changes remains a major challenge in glioblastoma (GBM) patients. In this work, we compared the performance of 2 different MR perfusion techniques, dynamic susceptibility contrast (DSC), and arterial spin labeling (ASL) to differentiate recurrent tumor and post-treatment changes from the volume of cellular tumor segmented from combined Deep Learning and multimodal MRI measurements, including multishell diffusion and perfusion. Methods: In this retrospective study, 137 MRIs from 107 patients with GBM were analyzed. Cellular tumor maps were segmented by 2 radiologists based on imaging, clinical history, and pathology. Multimodal MRI with perfusion and multishell diffusion were inputted into 5 nnU-Net Deep Learning models using either DSC or ASL with combination of multishell diffusion and standard MRI sequences to segment cellular tumor. Models with DSC and ASL were compared using segmentation performance (Dice score) and accuracy to detect recurrent tumor from post-treatment changes (area under the curve [AUC] under the receiver operating characteristic curve). Results: Segmentation performances were similar in both cases, with a median Dice score of 0.75 (IQR: 0.53-0.84) for ASL and 0.76 (IQR: 0.57-0.84). AUC was 0.88 (CI 0.82-0.94) for ASL and 0.86 (CI, 0.80-0.92) for DSC, and this difference was statistically significant (P < .05, n = 10 000 permutation test). In 11 individual cases, recurring disease was detected with ASL but missed with cerebral blood volume, including recurring tumor in the vicinity of a surgical cavity (n = 5), close to the skull base (n = 1), and adjacent to an Ommaya reservoir (n = 2). Conclusions: Our results demonstrate the utility of ASL in regions where susceptibility artifacts decrease the quality of DSC images.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To develop and validate a deep learning (DL) method to detect and segment enhancing and nonenhancing cellular tumor on pre- and posttreatment MRI scans of patients with glioblastoma and to predict overall survival (OS) and progression-free survival (PFS). Materials and Methods This retrospective study included 1397 MRIs in 1297 patients with glioblastoma, including an internal cohort of 243 MRIs (January 2010-June 2022) for model training and cross-validation and four external test cohorts. Cellular tumor maps were segmented by two radiologists based on imaging, clinical history, and pathology. Multimodal MRI with perfusion and multishell diffusion imaging were inputted into a nnU-Net DL model to segment cellular tumor. Segmentation performance (Dice score) and performance in detecting recurrent tumor from posttreatment changes (area under the receiver operating characteristic curve [AUC]) were quantified. Model performance in predicting OS and PFS was assessed using Cox multivariable analysis. Results A cohort of 178 patients (mean age, 56 years ± [SD]13; 121 male, 57 female) with 243 MRI timepoints, as well as four external datasets with 55, 70, 610 and 419 MRI timepoints, respectively, were evaluated. The median Dice score was 0.79 (IQR:0.53-0.89) and the AUC for detecting residual/recurrent tumor was 0.84 (95% CI:0.79- 0.89). In the internal test set, estimated cellular tumor volume was significantly associated with OS (hazard ratio [HR] = 1.04/mL, P < .001) and PFS (HR = 1.04/mL, P < .001) when adjusting for age, sex and gross total resection status. In the external test sets, estimated cellular tumor volume was significantly associated with OS (HR = 1.01/mL, P < .001) when adjusting for age, sex and gross total resection status. Conclusion A DL model incorporating advanced imaging could accurately segment enhancing and nonenhancing cellular tumor, classify recurrent/residual tumor from posttreatment changes, and predict OS and PFS in patients with glioblastoma. ©RSNA, 2024.

BOLD fMRI arises from a physiological and physical cascade of events taking place at the level of the cortical microvasculature which constitutes a medium with complex geometry. Several analytical models of the BOLD contrast have been developed but these have not been compared directly against detailed bottom-up modeling methods. Using a 3D modeling method based on experimentally measured images of mice microvasculature and Monte Carlo simulations, we quantified the accuracy of two analytical models to predict the amplitude of the BOLD response from 1.5T to 7T, for different TE and for both gradient echo and spin echo acquisition protocols. We also showed that accounting for the tridimensional structure of the microvasculature results in more accurate prediction of the BOLD amplitude, even if the values for SO ₂ were averaged across individual vascular compartments. A secondary finding is that modeling the venous compartment as two individual compartments results in more accurate prediction of the BOLD amplitude compared to standard homogenous venous modeling, arising from the bimodal distribution of venous SO ₂ across the microvasculature in our data.