Background and Purpose

Cavernous malformation (CM) in brain is one of the common neurovascular diseases. Most people with CM may experience serious signs and symptoms such as hemisensory deficits, cranial nerve deficits, hemiparesis, headaches, seizures, dizziness, and cerebral hemorrhage ^{1, 2}. Gamma Knife radiosurgery (GKRS) is a viable alternative treatment option for patients with surgically-inaccessible CMs or significant medical comorbidities. The primary therapeutic goal of GKRS is to prevent re-hemorrhage of the lesion. However, about 30% of patients still experience re-hemorrhage after GKRS ³. We aimed to identify early image biomarkers associated with future re-hemorrhage events after GKRS based on MRI radiomics. This study reported the longitudinal changes of MRI radiomic features in CM to observe the subtle changes of lesions during follow-ups to guide the patient management and follow-up monitoring.

Materials and Methods

We conducted a retrospective study of 162 CM patients treated by GKRS with overall 422 longitudinal imaging follow-ups within 4 years after GKRS, including 72 CMs with re-hemorrhage after GKRS (the hemorrhagic group) and the other 127 CMs without re-hemorrhage after treatment (the non-hemorrhagic group). Serial magnetic resonance images (MRI), including full series of T2-weighted images (T2W) and contrast-enhanced T1-weighted (CET1), were collected for each patient before GKRS and yearly follow-up thereafter from Taipei Veteran General Hospital between 1993 and 2020. The study was approved by the local Institutional Review Board.
Several postprocessing steps on the MRI were applied to reduce the discrepancy of imaging parameters using our previously published MRI radiomics platform (MRP, http://www.ym.edu.tw/~cflu/MRP_MLinglioma.html) with a graphic user interface built on MATLAB programming environment ⁴. The adjustment of image resolution will be first performed to resample all voxel size to isotropic spatial resolution (1.00 x 1.00 x 1.00 mm3) for each MRI contrast according to Image biomarker standardisation initiative (IBSI) ⁵. The CM lesions were manually delineated at each time point followed by the extraction of 1178 radiomic features (including the gray-scale intensity patterns, lesion shape/size geometry, and texture features).
Subjects were divided into two subsets based on the re-hemorrhage time points. The first subset included bleeding in the first year after GKRS, and the second subset included bleeding between the second and fourth years after GKRS. For the first subset, we collected complete information of hemorrhage time point (HT) and one year before hemorrhage (BH1). For the second subset, we collected complete information of HT, BH1, and two years before hemorrhage (BH2). In order to reveal the time effects and group effects, both subsets were tested by the paired t-test (time effects) and two-sample t-test (group effects) separately. We first applied a paired t-test to identify the radiomic features with significant changes before re-hemorrhage events (p<0.05). A two-sample t-test was then used to determine the difference of radiomic features between hemorrhagic and non-hemorrhagic groups for selecting radiomic features as early biomarkers (p<0.05). The processing workflow is shown in Figure 1.

Results and Discussion

Table 1 lists the clinical characteristics of 162 enrolled patients with 199 CM lesions. Results of statistical analysis showed that no significant difference (p>0.05) in age, gender, size of CCM, clinical symptoms, average max dose, and average isodose level between the re-hemorrhage and hemorrhage-free group. Few clinical characteristics, including the lesion location and the average margin dose of GKRS, showed significant differences (p<0.05) between the two groups.
For the re-hemorrhage at the first year after GKRS (the first subset), we identified six MRI radiomics features with significant differences (p<0.05 in Table 2). For the re-hemorrhage among the second to fourth year after GKRS (the second subset), we identified twenty MRI radiomics features with significant differences (p<0.05 in Table 2).
We further used these radiomic features to forecast the re-hemorrhage of CM before the bleeding occurrence. We found all of the selected radiomic features were texture features. Therefore, the conventional visual inspection without quantitative analysis based on radiomics would be difficult to find changes in lesion inhomogeneity. The changed score of the re-hemorrhagic group had a wide margin than the hemorrhagic-free group. This phenomenon indicated that the re-hemorrhagic lesions were relatively unstable.
For the first subset, the significant differences in raw BH1 scores between the two groups could warn us about future re-hemorrhage (Figure 2). For the second subset, most of the significant features showed a progressively increase or decrease during the two years before bleeding, only a few features showed an increase (p<0.05) followed by a slight decrease (Figure 3). Although the decreases were not statistically significant (p>0.05), the changes between BH2 and HT were increasing (p<0.05). However, the slight decrease in these features is still a cause for attention, and we expect that the dynamic changes of CM before re-hemorrhage would be further investigated in future studies. The significant differences in the changed scores among BH1 and BH2 could warn us about the bleeding that might occur in the future, once the scores increase or decrease beyond the normal ranges.

Conclusions

This study investigated the longitudinal changes of MRI radiomics features in patients with CM after receiving GKRS treatment. The observation of longitudinal changes of MRI radiomic features before re-hemorrhage could be helpful to identify early changes before bleeding and be beneficial to monitor the re-hemorrhages after GKRS.

Acknowledgements

This study was supported by Ministry of Science and Technology of Taiwan (109-2314-B-010-022-MY3), Taipei Veterans General Hospital, National Yang Ming Chiao Tung University, and University System of Taiwan (VGHUST110-G7-2-2).