Radiation therapy has developed in recent years, and many new techniques have emerged [11]. Techniques capable of increasingly accurate tumor localization have been developed to minimize the exposure of the normal brain to high radiation [12]. Radiation therapy is widely used for treating meningiomas that are difficult to access surgically, such as those in the skull base and optic pathway [13]. There are reports that radiation therapy for local control after subtotal excision can improve the 5-year progression-free survival by up to 95% [13]. Radiation therapy after subtotal excision is effective enough. A study tracked 84 meningiomas treated with stereotactic radiosurgery; the tumor volume was reduced by 33% after 24 months and 36% after 36 months [14, 15]. In our study, as a result of follow-up for approximately 3 years, the size and maximal diameter increased significantly in the observation group. However, growth was not evident or slightly decreased in the radiation treatment group. There was a significant difference in meningioma size between both groups. Conversely, there was no significant difference in the SI and contrast enhancement of meningioma between both groups after radiotherapy. This result is considered meaningful.
Both groups differed significantly regarding age and meningioma location. Patients in the radiation therapy group were younger, and meningiomas in the observation group were mostly located in the parasagittal or falx cerebri. Our study showed that radiation therapy was used more frequently in treating meningiomas that are difficult to access surgically. Examples are those located in the cerebellopontine angle, cerebellar tentorium, cavernous sinus, posterior fossa, olfactory bulb, sella, straight sinus, and optic sheath. Significantly more findings, such as CSF cleft, mass effect, bony invasion, necrosis/hemorrhage, venous sinus invasion, and recurrence, were observed in the radiation treatment group than in the observation group.
In this study, 2 patients in the radiation treatment group increased rapidly in tumor size even after radiotherapy. After surgery, the meningioma was confirmed as atypical. The possibility of atypical or malignant meningioma cannot be completely ruled out if there is a rapid increase in size after radiation therapy[16]. To date, no paper has studied the discrimination of radiomic features after radiation treatment of meningioma based on MRI. This study is the first to propose a radiology model based on brain MRI.
In our study, we could not extract meaningful radiomic features from T2WI and FLAIR, but we extracted radiomics of T1WI, CE-FLAIR, CE-T1WI, and five image combinations. A total of 24 radiomic features showed a high correlation with radiation treatment, and the combination of images showed good predictive power in training (AUC: 0.76) and validation (AUC: 0.79). The single images T1WI, CE-FLAIR, and CE-T1WI showed relatively high predictive power comparable to the combination of images.
We investigated the correlation between radiomic features and radiation therapy. Among 1595 radiomic features, 14 showed a high correlation in T1WI, CE-FLAIR, and CE-T1WI. Many of these features were textural features of the image, features measuring gray-level values, and radiomic features measuring heterogeneity in the texture patterns, local homogeneity, or heterogeneity of the image. These features showed the microscopic descriptions of the meningioma. These features were not easily discernible to the human eye or understood or interpretable in any particular meaning [17–21]. Here, the radiomic feature called CE-FLAIR_Small Dependence Low Gray Level Emphasis was a meaningful element commonly derived from univariate and multivariate analysis results. Radiomic features emphasizing gray level were significantly associated with brain invasion of meningiomas [17]. CE-T1WI original_shape_Compactness2, orginal_shape_Spherical Disproportion, and logarithm-glasm small area emphasis seemed to show changes in tumor size and shape. The values of these features were lower in the radiation treatment group than in the observation group, suggesting that changes in meningioma size and shape are related to radiation therapy. A study that investigated the association between radiomic and semantic features of meningioma reported that spherical disproportion was related to mass effect, speculation, and bony and venous sinus invasion [1].
Kurtosis, a radiomic feature commonly seen in T1WI and CE-T1WI, is a measure of the peakedness of the distribution of values in meningioma ROI, meaning that higher values do not converge to the mean but spread toward the tail of a normal distribution [22]. Radiomic features reflect the microscopic heterogeneity within tumors associated with radiation therapy. They are a new tool for predicting meningioma changes after radiation therapy. It is relatively easy to compare the change in SI of the image by measuring the ROI. However, it was a very interesting task as it provided information on the changes in the radiomic features of meningioma after radiation treatment.
Our results suggest that the radiomic features of meningiomas can be used to predict changes and after radiotherapy. This study attempted to extract radiomic features from each image by selecting T1WI, T2WI, FLAIR, CE-FLAIR, and CE-T1WI. CE-T1WI is commonly used to define macroscopic tumor boundaries and evaluate the extent of tumor invasion and blood supply [23]. T2 imaging is sensitive to watery tissues and can be used to detect the presence of edema [24]. Other studies have introduced features related to meningioma and brain invasion in T2WI [17].
However, this study showed no significant difference between the observation and treatment groups in the SI ratio in T2WI and FLAIR. Radiomic features also had no meaningful extracted information. This seems to require more in-depth research with more patients. In this study, the multiple sequence models (T1WI, T2WI, FLAIR, CE-T1WI, and CE-FLAIR combined) showed better predictive power than the single models. These results suggest that multiple sequences may provide more information about the tumor and better show radiation-related changes in meningiomas.
Our study had several limitations. First, as this was a retrospective study, histological confirmation was not performed in all patients. Meningioma removal surgery was performed on a limited basis only in some cases with enlarged tumors. Although evaluated using imaging, image evaluation may be subjective. Second, this study was conducted at a single institution with a limited sample size. Further studies with larger sample sizes from multiple centers are required for external validation. External validation is needed to evaluate the reproducibility and optimize the model. Third, there may be an unavoidable selection bias, as approximately 18% of the patients were excluded from the training and validation cohorts for various reasons. Fourth, MRI scans were retrospectively collected on three MRI machines with different devices and acquisition parameters, and radiomic features are sensitive to parameters. Therefore, it is necessary to normalize the MR images to obtain a standard normal distribution of image intensities. Finally, in the T1WI, T2WI, and FLAIR sequences, the meningioma and surrounding borders were unclear in some cases. We referenced the CE-T1WI sequence for visual guidance; however, deviations remained. In the future, multimodal studies, such as DWI and ADC map sequences, can be integrated into the model to further improve accuracy.
In conclusion, this study evaluated radiomic features and radiologic appearances of meningiomas, which were significantly different after radiotherapy. Our findings can help predict size reduction after radiotherapy for meningioma. A radiomic model using MR images can be useful as a biomarker to predict changes in meningiomas after radiation treatment. This is expected to have a positive effect on patient treatment.