Prognostic Value of Weekly Delta-Radiomics during MR-Linac Radiotherapy of Glioblastoma

PURPOSE/OBJECTIVE(S)

MRI after chemoradiotherapy (chemoRT) shows areas of presumed tumor growth in ≤ 50% of glioblastoma (GBM) patients, which can be true progression (TP) - tumor growth with poor treatment response, or pseudoprogression (PP) - edema and tumor necrosis with favorable treatment response. Patients with TP have median overall survival (OS) of only 7 months, while patients with PP have median OS of 36 months. However, on imaging, TP and PP are usually not discernible during treatment, making it difficult to adapt radiation for poor responders. The purpose of this study was to investigate the prognostic value of delta radiomic features from MR-Linac for GBM.

MATERIALS/METHODS

Using an IRB-approved prospective cohort of GBM patients undergoing 30 fractions of chemoRT to 60 Gy on a 0.35T MR-Linac, 2 regions of interest (ROI) were contoured on daily T2-weighted treatment set-up scans: 1) tumor/edema (lesion) and 2) post-surgical resection cavity (RC). The lesion ROI were used to calculate texture features: second order radiomics features based on the gray-level co-occurrence matrix (GLCM), gray-level size zone matrix (GLSZM), gray-level run length matrix (GLRLM), and neighborhood gray-tone difference matrix (NGTDM). Each of these describe the probability of spatial relationships of gray levels occurring within the ROI. Features from fraction 1 (pre-radiation) were subtracted from fractions 5, 10, 15, 25, and 30 to create delta features at 5 timepoints (D5-D30). Patient response was retrospectively defined as no progression (NP), TP, or PP. Supervised machine learning was utilized using a 500-tree random forest (RF) classification model with TP or PP as the outcome. Variable importance analysis was conducted by calculating the out-of-bag errors with multiple bootstrapped data sets. The most prognostic features were selected using the RF importance scores.

RESULTS

Thirty-six patients were screened for inclusion: 9 were excluded due to no T2 lesion (RC ROI only). Of the remaining 27 patients: 10 had NP, 11 had TP, and 6 had PP. Thirty-nine texture features, plus lesion volume and mean lesion intensity (for a total of 41 variables per time point) were calculated and included in the model. Of the 10 most prognostic features, 6 were from D10, suggesting that prognostic changes in the underlying lesion microenvironment are occurring within the first 10 fractions of treatment. The model selected GLSZM high gray-level zone emphasis (HGZE) D10, IBSI code 5GN9, as the most prognostic feature. The receiver operator characteristic (ROC) area under the curve (AUC) for GLSZM HGZE D10 was 0.94 (95% CI = 0.81-1.00).

CONCLUSION

Delta radiomic features extracted from MR-Linac imaging may predict between PP and TP in GBM patients during treatment, which is earlier than current methods. This could allow physicians to adapt/intensify treatment in real time for poorly responding patients. Future directions include analysis with a larger patient cohort and with additional MRI contrasts (MR-Linac multiparametric MRI).

Spectroscopic MRI Detects Occult Glioblastoma Invasion during Chemoradiation

PURPOSE/OBJECTIVE(S)

The standard of care for glioblastoma (GBM) includes surgical resection followed by adjuvant chemoradiation (chemoRT). Treatment margins are controversial since conventional imaging does not define the extent of infiltrating tumor cells. Whole-brain spectroscopic MRI (sMRI) allows for visualization of native metabolites in normal brain and tumor cells, and the relative choline to N-acetyl-aspartate ratio greater than 2 (rChoNAA>2) strongly correlates with the presence of occult GBM cells in otherwise normal-appearing brain. With an MRI-Linac, we are performing studies of adaptive radiotherapy to measure changes in cavity size, edema, and enhancement during chemoRT. We questioned whether rChoNAA>2 would change along with anatomical changes to inform clinical target volumes for adaptive chemoRT trials.

MATERIALS/METHODS

In a prospective study, 18 patients with primary GBM underwent daily MRI-guided chemoRT with standalone 3T sMRI generation of rChoNAA>2 maps at three timepoints before, during, and after chemoradiation. Conventional treatment volumes of T1 post-contrast and cavity (GTV2, i.e., boost) with or without FLAIR hyperintensity (GTV1) were compared to rChoNAA>2 volumes. DICE similarity coefficients were calculated to assess the spatial similarity of these volumes. Hausdorff distances were calculated to identify rChoNAA>2 extending beyond GTVs throughout the course of chemoradiation.

RESULTS

The mean GTV1 was 58.1 cc (range 0-251.4 cc), the mean GTV2 was 47.9 cc (range 0-139.9 cc), and the mean rChoNAA>2 volume was 31.1 cc (range 0-103.2 cc). rChoNAA>2 volumes did not significantly change over the course of chemoRT or correlate with measurement timepoint. The mean DICE similarity coefficient between GTV1 and rChoNAA>2 volumes was 0.39 (range 0-0.80), and the mean DICE similarity coefficient between GTV2 and rChoNAA>2 volumes was 0.29 (range 0-0.77). DICE similarity coefficients were significantly different from unity indicating spatial differences between rChoNAA>2 and conventional MRI volumes. The mean Hausdorff distances of rChoNAA>2 extending beyond GTV1 was 1.3 cm (range 0.7-2.1 cm), and the mean Hausdorff distances of rChoNAA>2 extending beyond GTV2 was 1.9 cm (range 0.8-2.9 cm), suggesting high-risk disease invading beyond what is visible on conventional MRI sequences.

CONCLUSION

Whole-brain sMRI with generation of rChoNAA>2 maps suggest conventional MRI does not fully capture the extent of disease in primary GBM throughout the course of chemoradiation. rChoNAA>2 maps often extend up to approximately 2 cm beyond conventional boost radiotherapy volumes. Further studies are ongoing to determine how sMRI can be used to adapt radiation target volumes during chemoradiation and escalate dose to occult disease.

Pattern Analysis of Daily Lesion Volume Trajectories for Early Prediction of Glioblastoma Progression During MR-Linac Radiotherapy

PURPOSE/OBJECTIVE(S)

Distinguishing between true progression (TP) and pseudoprogression (PP) post-radiotherapy (RT) is of paramount importance for treatment management of patients with glioblastoma (GBM). MR-Linac systems allow for daily monitoring of tumor changes throughout the course of RT. We hypothesized that the patterns of tumor volume change during RT may enable early prediction of treatment response.

MATERIALS/METHODS

Using an IRB-approved prospective cohort of GBM patients undergoing 30 fractions of chemoRT to 60 Gy on a 0.35T MR-Linac, tumor/edema (tumor lesion) regions of interest (ROI) were contoured on daily T2-weighted treatment set-up scans. The obtained tumor lesion (TL) volume changes during treatment were smoothed with a moving average Gaussian window over time. Non-negative Matrix Factorization (NMF) was applied to the data matrix D (N x F), containing the trajectories in its rows for each patient, where N is the number of patients analyzed and F is the number of fractions. NMF represents D as a linear combination of three temporal (hidden) patterns and their weights in each individual trajectory. The same analysis was performed for ΔD, containing the changes in volumes with reference to the first fraction. The calculated weights were scaled in [0, 1], expressed as probabilities (by ℓ1-normalization) and used as features in Linear Discriminant Analysis (LDA). The LDA model was trained to differentiate between no progression (NP), PP and TP, and assessed by leave-one-subject-out cross-validation.

RESULTS

Thirty-six patients were screened for inclusion: 9 were excluded due to no T2 lesion (resection cavity only). Of the remaining 27 GBM patients analyzed, 10 had no tumor growth on first post-RT diagnostic MRI, 6 were determined to have PP based on regression or long-term stability of findings, and 11 had TP due to continued progression of disease past 6 months, rapid patient death from disease, or tissue sampling showing active malignancy. With the use of only 2 features, LDA achieved an overall accuracy of 70.4% classifying correctly: 6 (60%), 4 (67%), and 9 (82%) patients with NP, PP, and TP, respectively. The temporal NMF patterns (monotonous decrease, rapid increase during the third part of the treatment, etc.) indicate that there is enough signal to classify patients' response based on the pattern tumor volume changes during RT.

CONCLUSION

We identified tumor dynamics' patterns during RT, indicative of differential behavior of tumor growth between TP and PP. Although with a limited number of patients, these initial results suggest that tumor volume changes during treatment may provide early markers of treatment response. This could allow physicians to adapt/intensify treatment in real time for poorly responding patients. Next steps include automating the process of real-time tumor volume monitoring by incorporating a deep learning solution for automatic volume delineation on daily treatment set-up scans.