Dysembryoplastic neuroepithelial tumors (DNETs) are classified as low-grade glioneuronal tumors (WHO grade I), which are rare and commonly benign.1 DNET predominantly occurs in the temporal lobe and affects children and adolescents, usually manifesting as intractable partial complex seizures and drug-resistant epilepsy.1–5 Given the benign nature, rare recurrence, and indolent course of DNETs, gross-total resection (GTR) is usually curative, without adjuvant chemotherapy or radiation.2,5–7
Resection is currently considered the only factor related to seizure outcome in most studies8,9 in which the prognosis of DNET has been addressed. However, recent studies have shown recurrence and malignant transformation following subtotal resection (STR) and even GTR in the long-term follow-up, which suggests that DNET may not be as benign as previously thought.10–12 In addition, some authors have reported that there may be some atypical DNETs that differ from the typical tumors in location, behavior, and characteristics.13,14 In order to have a more updated and deeper understanding of DNET prognosis, in our institution we introduced a new research area, radiomics, to elucidate MR images in DNET patients. Via high-throughput computing with the application of machine learning and artificial intelligence for medical image processing and analysis, MR images can be converted into quantitative radiomics data15–17 that may be useful to neurosurgeons, leading to increased accuracy of prognosis evaluation and improved treatment strategies for DNET patients.16–18
The radiomics features of DNET have, to our knowledge, not previously been explored systematically. To make the diagnosis and resection of DNET more accurate and effective, it is necessary to carry out a more in-depth study. In our research, we aimed to perform a preliminary summary of DNET prognostic factors in patients with postoperative seizures by using radiomics analysis to more accurately differentiate lesions (satellite lesions) that may lead to recurrent epilepsy in regions with edema. We also evaluated the radiomics features of both the tumor and edema regions of DNET patients to more accurately determine tumor boundaries and thus achieve more complete resection. To reduce the probability of postoperative seizure recurrence, receiver operating characteristic (ROC) curve analysis of the above radiomics features was used to explore the predictive values of the edema and tumor regions for postoperative recurrent seizure. Finally, a systematic review following the PRISMA process was used to locate additional studies that used radiomics in the diagnosis or prediction of epilepsy. We selected original articles for analysis of the direct function and ability of radiomics to detect postoperative seizures, including the purpose, inclusion of related diseases, and advantages and disadvantages of the selected literature. Study reports that did not include descriptions of standard radiomics procedures were not considered for inclusion.
Methods
Search Method
We searched for basic and clinical studies published in the PubMed and Web of Science databases during the period from January 1, 2018, to December 29, 2021. The following terms were queried singly and/or in combination: radiomics, epilepsy, and seizure. Studies that did not meet selection criteria were excluded from this review. The remaining studies were carefully checked for eligibility for inclusion in accordance with PRISMA guidelines (Fig. 1). Studies were included if they reported the detection of epilepsy using radiomics and were published as full-text journal articles in English. Exclusion criteria were as follows: 1) review rather than a report of original data; 2) not about epilepsy or seizure; 3) not about diagnosis or prediction of epilepsy; and 4) method did not include radiomics. Two independent reviewers screened and collected data from each report. Titles, abstracts, and methods among these reports were screened by both reviewers. All excluded studies were documented with reasons for exclusion. Any disagreement was resolved through consensus. Assessment of studies was performed to identify and exclude those with a lack of information related to radiomics and epilepsy, meta-analysis, data synthesis, or bias analysis within the studies.
Flowchart of the literature search. This figure shows our PRISMA process of searching the literature related to radiomics and epilepsy.
Participant Selection
Ethics committee approval was acquired, and patient informed consent was waived because this study was retrospective. We retrospectively collected data from 18 patients with surgically and histologically confirmed DNET who were treated at Nanfang Hospital during the period from June 2011 to July 2017. The inclusion criteria for these patients were as follows: 1) histologically confirmed DNET,1 2) preoperative MR images available, and 3) previous GTR of DNET. Patients with one of the following conditions were excluded: 1) diagnosis with other diseases leading to epilepsy, 2) other lesions that could not be diagnosed as DNET that appeared on preoperative MR images, and 3) missing postoperative follow-up records. After consideration of these criteria, 18 patients were selected for our retrospective study. According to the follow-up results, the patients were divided into two groups: the epilepsy recurrence group (ERG) and the epilepsy nonrecurrence group (ENRG). The postoperative follow-up period for these patients ranged from 3 to 108 months. The pathological results of the tumors were reviewed and compiled by two pathologists who were blinded to the radiographic results. The diagnosis of DNET was established based on the 2016 WHO Classification of Tumors of the Central Nervous System.1 The basic information for the study patients is shown in Table 1.
Patient clinical data
Pt No. | Sex | Age at Op (yrs) | Symptoms | Tumor Location | |
---|---|---|---|---|---|
Initial | Duration (mos) | ||||
1 | F | 48 | Dizziness | 3 | Temporal, rt |
2 | F | 5 | Complex partial seizure, dizziness, visual disturbances | 6 | Parietofrontal, lt |
3 | F | 28 | Complex partial seizure | 1 | Frontal, lt |
4 | F | 26 | Complex partial seizure | 72 | Frontal, lt |
5 | F | 11 | Fatigue | 36 | Brainstem |
6 | F | 19 | Complex partial seizure, consciousness disorders | NA | Temporal, lt |
7 | M | 11 | Complex partial seizure | 0.5 | Parieto-occipital |
8 | M | 12 | Headache, dizziness, fatigue | 4 | Pineal gland |
9 | M | 23 | Complex partial seizure, unconsciousness | 3 | Sellar region |
10 | M | 16 | Complex partial seizure, unconsciousness | 3 | Apical, rt |
11 | M | 30 | Complex partial seizure, dizziness | 0.5 | Temporal, rt |
12 | M | 30 | Complex partial seizure, dizziness | 72 | Frontal, lt |
13 | M | 5 | Hearing loss | 19 | Cerebellum, rt |
14 | F | 45 | Complex partial seizure, headache, dizziness, nausea | 36 | Tectum, mesencephalic |
15 | F | 6 | Complex partial seizure | 3 | Frontal, rt |
16 | F | 23 | Complex partial seizure, unconsciousness | 8 | Temporal |
17 | M | 2 | Complex partial seizure | 25 | Hypothalamus |
18 | F | 22 | Complex partial seizure | 60 | Temporal, lt |
Mean | 20.1 | 20.7 | |||
Range | 2–48 | 0.5–72 |
NA = not available; pt = patient.
MRI Acquisition and Image Processing
All patients included in the study had undergone resection and had preoperative MR images. All MR images of all patients were obtained with the same MRI scanner (Discovery ST, GE Healthcare) according to the standard clinical scanning protocols at our hospital. Two types of MRI sequences were selected for the study: 1) T1-weighted spin-echo images (T1WI) with a repetition time/echo time (TR/TE) of 2283.9/24.3 msec, and 2) FLAIR with a TR/TE of 7502.0/122.1 msec. The distribution, extension, shape, and signal features of the lesions were evaluated. These included first-order features, shape features, and gray-level co-occurrence matrix (GLCM) features, gray-level size zone matrix (GLSZM), and gray-level run-length matrix (GLRLM) features. Two types of MRI sequences, T1-weighted and FLAIR, were selected for the radiomics study. T1-weighted images were used to identify tumors, and FLAIR images were used to identify edema regions. Both regions were delineated by an experienced neurosurgeon.
For the segmentation, we used 3D Slicer 5.2.0 (http://www.slicer.org) operating on Mac OS X. The tumor and edema regions were manually mapped on 3D Slicer by an experienced neurosurgeon.17,19 With the use of 3D Slicer, the two MR images of each patient, T1-weighted and FLAIR, were located on the same slice automatically (Fig. 2).
Schematic diagram of radiomics feature extraction. The tumor and edema regions are scanned by MRI, and the acquired images are automatically segmented into voxels of the same size. After measurement of different dimensions, the dimensional data of each voxel will be separated and displayed in the form of radiomics data.
Radiomics Feature Extraction and Selection
A set of 94 radiomics features were extracted from the tumor and peritumoral edema regions (edema in the surrounding tissue) through the use of 3D Slicer Radiomics running on Mac OS X. Features of the tumor region were extracted from T1-weighted images, and features of the edema region were extracted from FLAIR images. Subsequently, the Mann-Whitney U-test was used to select all extracted radiomics features with significant differences (p < 0.05). This difference was based on whether epilepsy was recurrent.
Statistical Analysis
Our process of statistical analysis included screening, comparison, and inspection. The Mann-Whitney U-test was used in the screening process, the Friedman test was used in the comparison process, and the ROC curve and area under the curve (AUC) were used in the inspection for evaluation.
The Mann-Whitney U-test was used in selection. In order to clarify the difference between the edema region and tumor region in each group, the Friedman test was used for feature analysis in group comparisons and Bonferroni correction was used for the significance value. The selected features were converted to normalization by Min-Max normalization to eliminate the effect of dimension. The ROC curve was used to measure the predictive ability of radiomics features associated with postoperative recurrent epilepsy.
All data were analyzed using IBM SPSS version 26.0 (IBM Corp.). All statistical tests used in this study were two-sided, and statistical significance was defined as p < 0.05.
Results
Clinical Features
This study included 18 patients with DNET. The clinical and pathological information of DNET patients is presented in Table 1. Eight (44%) patients were male and 10 (56%) patients were female. The median age at diagnosis was 20.1 years (range 2–48 years). Most patients had the common clinical manifestations, such as epilepsy (74%) and dizziness (32%). Two (11%) patients suffered from visual disturbances, 4 (22%) patients had consciousness disturbance, and 1 (5%) patient presented with hearing impairment. The duration of clinical manifestations ranged from 0.5 to 72 months (mean 20.7 months). All patients accepted the treatment of GTR, among whom 17 patients had good recovery and 1 patient presented with a recurrent tumor after the initial surgery 9 years previously. The detailed data are presented in Table 1.
Search Result
A flow diagram of our study is shown in Fig. 1. There were 32 potentially suitable publications. After the removal of duplicates, 20 remaining papers were screened. Of these, 4 papers were excluded because they were not about epilepsy or seizure; 2 papers were excluded because they were reviews; 3 papers were excluded because they did not focus on diagnosis or predictions regarding epilepsy; and 2 papers were excluded because they lacked information about radiomics. According to the results of the literature search (Fig. 1), a total of 9 papers met the criteria, for which the time span was from January 1, 2018, to December 29, 2021.20–28
Radiomics Feature Selection
The Mann-Whitney U-test was conducted to select 11 features with significant differences (p < 0.05), namely gray-level nonuniformity (GLNU), run-length nonuniformity (RLNU), least axis (LA), major axis, maximum 2D diameter column (M2DDC), maximum 2D diameter row (M2DDR), maximum 2D diameter slice (M2DDS), maximum 3D diameter (M3DD), minor axis, surface area (SA), and volume, from three different types: GLRLM, GLSZM, and shape. These features showed significant differences between the two groups, as mentioned above. The features that are classified into the shape category include the LA, major axis, M2DDC, M2DDR, M2DDS, M3DD, minor axis, SA, and volume. The features classified as GLRLM and GLSZM include GLNU and RLNU. The detailed selection data are presented in Supplemental Table 1. It is worth noting that GLRLM and GLSZM are used to describe texture, whereas the other features are used to describe pixel shape. In radiomics, texture is defined as the gray scale of the gray-level zone as well as our region of interest (ROI), including the length and level of the gray scale. A gray-level zone is defined as the number of connected voxels that share the same gray-level intensity, an important index used in radiomics to evaluate tumor heterogeneity. In radiomics, shape and volume do not refer to the size of objects in reality but are just a measure of pixels. These features are independent from the gray-level intensity distribution in the ROI and are calculated on the nonderived image and mask. For a detailed explanation of the above radiomics features, see the PyRadiomics website (https://pyradiomics.readthedocs.io/en/latest/).
Radiomics Feature Analysis
According to our study, the mean values of texture features, which refer to GLRLM and GLSZM, are higher in the ERG than the ENRG (Fig. 3). For example, the mean value of the ERG was 1069 and of the ENRG was 570 in the GLNU of the edema region. The mean value of the ERG was 1644 and of the ENRG was 597 in the GLNU of the tumor region. Meanwhile, mean values from shape features are higher in the ERG than in the ENRG (Fig. 4). The Friedman test shows that values have no significant difference (p > 0.05) between the edema region and tumor region among the 11 features we have selected in each patient in both groups. However, the M2DDC and SA show significant differences (p < 0.05) between the edema region and tumor region in the ERG.
Texture features of prognosis prediction by Min-Max normalization. Texture features from the ERG and ENRG are transformed by Min-Max normalization. The mean value is presented.
Shape features of prognosis prediction by Min-Max normalization. Shape features from the ERG and ENRG are transformed by Min-Max normalization. The mean value is presented.
Performance of Radiomics Features
For the 11 radiomics features we selected, the AUC (in the ERG) exceeded 0.8 (Fig. 5). The highest AUC was for M3DD, a shape feature that reached 0.945. The RLNU and GLNU, as texture features in the edema region, reached 0.836. Compared with other radiomics features, those we selected have higher predictive ability (Table 2). Whether the predictive performance of these features is unsatisfactory is not known. However, similar studies on brain tumors have shown an AUC greater than 0.6, which is considered highly predictive.29 Based on those data, some of the radiomics features we report here, especially our selected features, may be accurate in predicting postoperative recurrent seizures in DNET patients.
AUC of selected radiomics features. The specificity and sensitivity of selected radiomics features are shown. The AUC represents the probability that the radiomics estimation has correctly predicted postoperative epilepsy recurrence.
AUC values of radiomics features
Feature | Area | SEM* | p Value† | Limit‡ | |
---|---|---|---|---|---|
Lower | Upper | ||||
GLNU edema | 0.818 | 0.108 | 0.047 | 0.606 | 1.000 |
GLNU tumor | 0.800 | 0.115 | 0.062 | 0.574 | 1.000 |
RLNU edema | 0.836 | 0.105 | 0.036 | 0.630 | 1.000 |
RLNU tumor | 0.873 | 0.089 | 0.020 | 0.698 | 1.000 |
GLNU tumor 2 | 0.873 | 0.089 | 0.020 | 0.698 | 1.000 |
LA edema | 0.818 | 0.108 | 0.047 | 0.607 | 1.000 |
LA tumor | 0.855 | 0.103 | 0.027 | 0.653 | 1.000 |
Major axis tumor | 0.873 | 0.098 | 0.020 | 0.681 | 1.000 |
M2DDC edema | 0.855 | 0.101 | 0.027 | 0.657 | 1.000 |
M2DDC tumor | 0.873 | 0.090 | 0.020 | 0.696 | 1.000 |
M2DDR tumor | 0.945 | 0.059 | 0.006 | 0.830 | 1.000 |
M2DDS tumor | 0.855 | 0.096 | 0.027 | 0.665 | 1.000 |
M3DD tumor | 0.945 | 0.059 | 0.006 | 0.830 | 1.000 |
Minor axis tumor | 0.891 | 0.084 | 0.015 | 0.727 | 1.000 |
SA edema | 0.800 | 0.111 | 0.062 | 0.581 | 1.000 |
SA tumor | 0.855 | 0.096 | 0.027 | 0.666 | 1.000 |
Volume tumor | 0.836 | 0.101 | 0.036 | 0.638 | 1.000 |
Assumes nonparametric.
Null hypothesis: true area = 0.5.
95% CI.
Discussion
In a previous study, DNET was considered to be a benign tumor that would not recur.30 Now DNET has become the second most common type of pediatric tumor that leads to chronic drug-resistant and partial complex seizures.2,8,31 Surgical treatment has become the best way to treat tumors, but the efficacy of adjuvant therapy such as radiotherapy and chemotherapy remains to be verified.2,8,32,33 Kim et al. reported a progressive form of DNET, which reminded us that DNET may not be as benign as we supposed.10 Postoperative recurrent seizure has become the most common manifestation of poor tumor prognosis.34–36 Although epilepsy is not the only postoperative complication of DNET, the absence of seizures is an indicator of the recovery of advanced neurological function after DNET. Therefore, most studies regard nonrecurrence of epilepsy after DNET as the most important prognostic indicator.3,5,6,8,34 The region of resection is one of the most significant factors leading to epilepsy recurrence. In many studies, GTR is the riskiest factor associated with postoperative recurrent seizure.5,32–35,37,38
Despite the importance of the adequacy of the GTR resection region, neurosurgeons hesitate to expand this region of resection.5,34,36 In the present study, we aimed to use radiomics as a decision-making method based on the postoperative recurrent seizure outcome. In our results, the performance of the selected radiomics features was satisfactory, and therefore this method may be valuable for the neurosurgeon in performing tumor region identification and resection. With this method, recurrence of postoperative seizures in DNET patients may no longer be uncertain, but can be quantified with radiomics techniques. Moreover, the radiomics findings in the present study suggest that greater atypia of tumor tissue is closely related to the prognosis of DNET.
Application of Radiomics in Epilepsy
We conducted this systematic review to emphasize that radiomics can be used to evaluate epilepsy more accurately. The application of radiomics to epilepsy treatment can be divided into two aspects: diagnosis and prediction of epilepsy. Recent studies have shown that radiomics methods can diagnose epilepsy by detecting tiny epilepsy lesions.22 A research team led by Jie Tian applied radiomics to predict postoperative epilepsy in low-grade glioma and showed that radiomics could well predict the recurrence of epilepsy on low-grade glioma by analyzing MRI.20,23 The accuracy of prediction by radiomics features exceeded 0.7. However, few published studies have reported the use of radiomics to assess or measure epilepsy.
Research regarding radiomics and epilepsy is still in the early stage and lacks sufficient data. The breakthrough of radiomics research in a variety of epilepsy-related diseases shows that it is not accidental that the electrophysiological phenomenon of epilepsy can also be evaluated by radiomics in some tumors. With the rapid progress of machine learning algorithms and artificial intelligence, more applicable and accurate models will be developed in the near future. For more details, see Supplemental Table 2.
Texture: Satellite Lesions in the Edema Region
According to our results, the boundaries of DNETs are not clear or cannot be recognized by human eyes under a microscope. In the present study, the texture features of the edema region and the tumor region were compared. In both the ERG and ENRG, the texture values from the tumor and edema regions did not have a statistically significant difference (p > 0.05) in each patient; therefore, our results indicated that the homogeneity between the edema and tumor regions is similar in each patient.
Given this radiomics homogeneity, can we use radiomics to accurately distinguish edema from tumor? Based on our results, the tumor boundary of DNET may be ambiguous, and in a case like this we support GTR or even extended resection of DNET. For any residual lesion that remains, the ambiguity between the edema and tumor regions may account for recurrent seizure in patients who have undergone surgery. In some cases, seizure occurs with tumor enlargement.10,39 After the resection, the lesion is revealed to include dysplastic floating neurons, which is one of the typical pathologies associated with DNET.8,14,39 With the use of radiomics features, heterogeneity of the edema region was significantly greater in the ERG than in the ENRG. This suggests that the cellular atypia in the edema region is greater, and there may be some lesions leading to the recurrence of epilepsy. These lesions may progress and cause postoperative recurrent seizures and, finally, even recurrent tumor. These lesions may be new epileptic lesions and cause abnormal discharges in the cerebral cortex but be invisible to human eyes aided by a microscope.
Other recent research has indicated that in DNET satellite lesions contribute most among the prognostic factors, even more than the influence of GTR. Satellite lesions are identified as small DNET lesions (pathologically confirmed) in the edema region around the tumor. The presence of satellite lesions not only influences seizures but also leads to progression and recurrence of tumors.13,38,40 Tumor tissues can spread to the edema region, and therefore, although DNETs are considered to be benign, these tumors are not stable. Spreading cancer tissues may form hidden small satellite lesions that may cause postoperative recurrent seizures or even a recurrent tumor. The pathological certainty that residual tumors can cause recurrent seizures has been confirmed in research on satellite lesions,38 and we have used radiomics to detect the same conditions as satellite lesions.
It is worth mentioning that the heterogeneity of the edema and tumor regions is the same in the ENRG. Compared with the ERG, the texture heterogeneity of the edema region in the ENRG is smaller. In essence, the heterogeneity of the edema region may be caused by the satellite lesions mentioned above. The regional heterogeneity of edema in the ERG is higher because satellite lesions have been formed and detected by radiomics. However, the ENRG is also very likely to have satellite lesions while it is still in the early stage, so this manifestation of heterogeneity is relatively small (in terms of radiomics). In the ERG, denser satellite lesions in the edema region lead to recurrent epilepsy. Therefore, we suggest that the formulation of an operation plan includes removal of as much of the edema region as possible.
In our study, the texture of the original tumor was an important correlative factor of postoperative recurrent seizure. Therefore, we speculate that the pathological type of tumor may be related to the prognosis. Our measurement of texture was performed mainly in two dimensions: grayscale and run length, which includes GLNU and RLNU. In the ERG, both the edema region and tumor region have lower homogeneity than the ENRG. This means that the cell atypia in the tumor region and peritumoral edema region is greater in the ERG than in the ENRG. Studies have shown that radiomics can distinguish low-grade and high-grade gliomas from primary brain tumors41 and that measurements of radiomics heterogeneity are associated with pathological outcomes of lesions, with greater heterogeneity suggesting more severe pathological damage.42,43 Such great differences in tumor heterogeneity between the two groups are attributable to different pathological types. Just as different MRI signs indicate different histopathological types,9 differences in radiomics outcomes indicate that different histopathological types may have an impact on the prognosis of DNET. However, relevant studies are still lacking. In addition, our results are different from those of some previous studies.44,45 Radiomics is more sensitive and can detect changes that are indistinguishable to the human eye, and previous studies have not supported the efficacy of tumor pathological classification for tumor prognosis. Because of our incomplete pathological evidence, further research is needed to confirm the full meaning of our observations.
Radiomics: Predictive Value of Features
According to our results, radiomics features of preoperative images have a certain predictive value for postoperative recurrent seizures in DNET patients. Among the 11 radiomics features we selected, their accuracy in predicting epilepsy recurrence reached more than 0.8. From the contrast of texture and shape features, shape features are more valuable in predicting postoperative recurrent seizure. Texture features mainly focus on two aspects, the similarity of the whole gray scale and the joint distribution of gray-scale size. Combined with the definition of radiomics features, tissue heterogeneity of the tumor and edema regions has predictive value for postoperative recurrent seizure.
It is worth noting that the edema region also has predictive significance for postoperative recurrent seizure, which is rare in previous studies. In other studies, the peritumoral edema region of central nervous system tumors has been found to be significantly heterogeneous and can be used to predict the site of tumor recurrence.46 Similarly, greater heterogeneity in the peritumoral edema region indicated greater cellular atypia in the edema region in our study. In the ERG, cellular atypia in the edema region progresses into dense satellite lesions. These satellite lesions eventually lead to the recurrence of epilepsy.
Study Limitations
There are some limitations of our study. First, our sample size is small, so some potential results may be overlooked and skewness is inevitable, which also increases the possibility of type I error in nonparametric data. Considering that our data are not normally distributed, the t-test is not applicable to our study. The existing literature suggests that the best method for small sample size research is a nonparametric method.47,48 Second, the lesions are mainly in the temporal lobe in our collected cases, which differs from collected cases in other reported studies. The pathological changes of the lesions may be different in different parts of the brain, resulting in differences in clinical, imaging, and histological manifestations. We were also unable to obtain case data for verification in public databases such as The Cancer Genome Atlas (TCGA) because of the rarity of DNET. More detailed pathological analysis is needed for improved study of the heterogeneity and composition of the tumor region in patients with DNET.
Conclusions
Regions of brain edema have an important predictive value for postoperative recurrent epileptic seizures because of the presence of satellite lesions that can be detected with radiomics. Although there are some shortcomings, the initial achievements with the use of radiomics in this study to identify brain regions associated with postoperative seizures indicate that radiomics may be a valuable tool for neurosurgeons in the evaluation of epilepsy.
Acknowledgments
This work was supported by grants from the Natural Science Fund of Guangdong Province (no. 2017A030313597), “Climbing Program” Special Fund of Guangdong Province (nos. pdjh2019b0100 and pdjh2020b0112), and Southern Medical University (nos. LX2016N006, KJ20161102, 201912121004S, and 201912121013).
Disclosures
The authors report no conflict of interest concerning the materials or methods used in this study or the findings specified in this paper.
Author Contributions
Conception and design: Wang. Acquisition of data: Luo. Analysis and interpretation of data: Wang, Luo. Drafting the article: Wang, Luo, Chen. Critically revising the article: Wang, Chen, Deng. Reviewed submitted version of manuscript: Wang, Luo, Chen, Deng. Statistical analysis: Wang, Luo, Chen, Long. Administrative/technical/material support: Long, Yang, Qi. Study supervision: Wang.
Supplemental Information
Online-Only Content
Supplemental material is available online.
Supplemental Tables. https://thejns.org/doi/suppl/10.3171/2022.7.FOCUS2254.
References
- 1↑
Louis DN, Perry A, Reifenberger G, et al. The 2016 World Health Organization Classification of Tumors of the Central Nervous System: a summary. Acta Neuropathol. 2016;131(6):803–820.
- 2↑
Pindrik J, Hoang N, Smith L, et al. Preoperative evaluation and surgical management of infants and toddlers with drug-resistant epilepsy. Neurosurg Focus. 2018;45(3):E3.
- 3↑
Zhang JG, Hu WZ, Zhao RJ, Kong LF. Dysembryoplastic neuroepithelial tumor: a clinical, neuroradiological, and pathological study of 15 cases. J Child Neurol. 2014;29(11):1441–1447.
- 4
Jenny B, Smoll N, El Hassani Y, et al. Pediatric epilepsy surgery: could age be a predictor of outcomes? J Neurosurg Pediatr. 2016;18(2):235–241.
- 5↑
Bonney PA, Boettcher LB, Conner AK, et al. Review of seizure outcomes after surgical resection of dysembryoplastic neuroepithelial tumors. J Neurooncol. 2016;126(1):1–10.
- 6↑
Nolan MA, Sakuta R, Chuang N, et al. Dysembryoplastic neuroepithelial tumors in childhood: long-term outcome and prognostic features. Neurology. 2004;62(12):2270–2276.
- 7↑
Radhakrishnan A, Abraham M, Vilanilam G, et al. Surgery for “Long-term epilepsy associated tumors (LEATs)”: seizure outcome and its predictors. Clin Neurol Neurosurg. 2016;141:98–105.
- 8↑
Luzzi S, Elia A, Del Maestro M, et al. Dysembryoplastic neuroepithelial tumors: what you need to know. World Neurosurg. 2019;127:255–265.
- 9↑
Chassoux F, Daumas-Duport C. Dysembryoplastic neuroepithelial tumors: where are we now? Epilepsia. 2013;54(suppl 9):129–134.
- 10↑
Kim AH, Thompson EA, Governale LS, et al. Recurrence after gross-total resection of low-grade pediatric brain tumors: the frequency and timing of postoperative imaging. J Neurosurg Pediatr. 2014;14(4):356–364.
- 11
Braoudaki M, Lambrou GI, Papadodima SA, Stefanaki K, Prodromou N, Kanavakis E. MicroRNA expression profiles in pediatric dysembryoplastic neuroepithelial tumors. Med Oncol. 2016;33(1):5.
- 12↑
Moazzam AA, Wagle N, Shiroishi MS. Malignant transformation of DNETs: a case report and literature review. Neuroreport. 2014;25(12):894–899.
- 13↑
Khan NI, Khanna L, Renjen PN, Fernandes CS. An unusual recurrence of dysembroplastic neuroepithelial tumours after a seizure-free period of 8 years. BMJ Case Rep. 2013;2013:bcr2013010469.
- 14↑
Daghistani R, Miller E, Kulkarni AV, Widjaja E. Atypical characteristics and behavior of dysembryoplastic neuroepithelial tumors. Neuroradiology. 2013;55(2):217–224.
- 15↑
Court LE, Fave X, Mackin D, Lee J, Yang J, Zhang L. Computational resources for radiomics. Transl Cancer Res. 2016;5(4):340–348.
- 16↑
Bernasconi A, Bernasconi N, Bernhardt BC, Schrader D. Advances in MRI for ‘cryptogenic’ epilepsies. Nat Rev Neurol. 2011;7(2):99–108.
- 17↑
Liu Z, Wang S, Dong D, et al. The applications of radiomics in precision diagnosis and treatment of oncology: opportunities and challenges. Theranostics. 2019;9(5):1303–1322.
- 18↑
Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. 2017;14(12):749–762.
- 19↑
Lambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48(4):441–446.
- 20↑
Mo J, Liu Z, Sun K, et al. Automated detection of hippocampal sclerosis using clinically empirical and radiomics features. Epilepsia. 2019;60(12):2519–2529.
- 21
Dai XJ, Liu H, Yang Y, Wang Y, Wan F. Brain network excitatory/inhibitory imbalance is a biomarker for drug-naive Rolandic epilepsy: a radiomics strategy. Epilepsia. 2021;62(10):2426–2438.
- 22↑
Cheong EN, Park JE, Jung DE, Shim WH. Extrahippocampal radiomics analysis can potentially identify laterality in patients with MRI-negative temporal lobe epilepsy. Front Neurol. 2021;12:706576.
- 23↑
Wang Y, Wei W, Liu Z, et al. Predicting the type of tumor-related epilepsy in patients with low-grade gliomas: a radiomics study. Front Oncol. 2020;10:235.
- 24
Zhang Y, Yan P, Liang F, Ma C, Liang S, Jiang C. Predictors of epilepsy presentation in unruptured brain arteriovenous malformations: a quantitative evaluation of location and radiomics features on T2-weighted imaging. World Neurosurg. 2019;125:e1008–e1015.
- 25
Xu J, Yu Y, Li Q, et al. Radiomic features as a risk factor for early postoperative seizure in patients with meningioma. Seizure. 2021;93:120–126.
- 26
Liu Z, Wang Y, Liu X, et al. Radiomics analysis allows for precise prediction of epilepsy in patients with low-grade gliomas. Neuroimage Clin. 2018;19:271–278.
- 27
Sun K, Liu Z, Li Y, et al. Radiomics analysis of postoperative epilepsy seizures in low-grade gliomas using preoperative MR images. Front Oncol. 2020;10:1096.
- 28↑
Park YW, Choi YS, Kim SE, et al. Radiomics features of hippocampal regions in magnetic resonance imaging can differentiate medial temporal lobe epilepsy patients from healthy controls. Sci Rep. 2020;10(1):19567.
- 29↑
Kniep HC, Madesta F, Schneider T, et al. Radiomics of brain MRI: utility in prediction of metastatic tumor type. Radiology. 2019;290(2):479–487.
- 30↑
Daumas-Duport C, Scheithauer BW, Chodkiewicz JP, Laws ER Jr, Vedrenne C. Dysembryoplastic neuroepithelial tumor: a surgically curable tumor of young patients with intractable partial seizures. Report of thirty-nine cases. Neurosurgery. 1988;23(5):545–556.
- 31↑
Wang J, Zhang Y, Zhang H, et al. Nucleus accumbens shell: a potential target for drug-resistant epilepsy with neuropsychiatric disorders. Epilepsy Res. 2020;164:106365.
- 32↑
Bergthold G, Bandopadhayay P, Bi WL, et al. Pediatric low-grade gliomas: how modern biology reshapes the clinical field. Biochim Biophys Acta. 2014;1845(2):294–307.
- 33↑
Chao L, Tao XB, Jun YK, Xia HH, Wan WK, Tao QS. Recurrence and histological evolution of dysembryoplastic neuroepithelial tumor: a case report and review of the literature. Oncol Lett. 2013;6(4):907–914.
- 34↑
Englot DJ, Berger MS, Barbaro NM, Chang EF. Factors associated with seizure freedom in the surgical resection of glioneuronal tumors. Epilepsia. 2012;53(1):51–57.
- 35↑
Robert-Boire V, Desnous B, Lortie A, et al. Seizures in pediatric patients with primary brain tumors. Pediatr Neurol. 2019;97:50–55.
- 36↑
Rydenhag B, Flink R, Malmgren K. Surgical outcomes in patients with epileptogenic tumours and cavernomas in Sweden: good seizure control but late referrals. J Neurol Neurosurg Psychiatry. 2013;84(1):49–53.
- 37↑
Devaux B, Chassoux F, Landré E, et al. Surgery for dysembryoplastic neuroepithelial tumors and gangliogliomas in eloquent areas. Functional results and seizure control. Neurochirurgie. 2017;63(3):227–234.
- 38↑
Slegers RJ, Blumcke I. Low-grade developmental and epilepsy associated brain tumors: a critical update 2020. Acta Neuropathol Commun. 2020;8(1):27.
- 39↑
Alexander H, Tannenburg A, Walker DG, Coyne T. Progressive dysembryoplastic neuroepithelial tumour. J Clin Neurosci. 2015;22(1):221–224.
- 40↑
Yang J, Kim SK, Kim KJ, et al. Satellite lesions of DNET: implications for seizure and tumor control after resection. J Neurooncol. 2019;143(3):437–445.
- 41↑
Haubold J, Demircioglu A, Gratz M, et al. Non-invasive tumor decoding and phenotyping of cerebral gliomas utilizing multiparametric 18F-FET PET-MRI and MR fingerprinting. Eur J Nucl Med Mol Imaging. 2020;47(6):1435–1445.
- 42↑
Liu Y, Dong D, Zhang L, et al. Radiomics in multiple sclerosis and neuromyelitis optica spectrum disorder. Eur Radiol. 2019;29(9):4670–4677.
- 43↑
Braman NM, Etesami M, Prasanna P, et al. Intratumoral and peritumoral radiomics for the pretreatment prediction of pathological complete response to neoadjuvant chemotherapy based on breast DCE-MRI. Breast Cancer Res. 2017;19(1):57.
- 44↑
Minkin K, Klein O, Mancini J, Lena G. Surgical strategies and seizure control in pediatric patients with dysembryoplastic neuroepithelial tumors: a single-institution experience. J Neurosurg Pediatr. 2008;1(3):206–210.
- 45↑
Honavar M, Janota I, Polkey CE. Histological heterogeneity of dysembryoplastic neuroepithelial tumour: identification and differential diagnosis in a series of 74 cases. Histopathology. 1999;34(4):342–356.
- 46↑
Long H, Zhang P, Bi Y, et al. MRI radiomic features of peritumoral edema may predict the recurrence sites of glioblastoma multiforme. SSRN. Preprint posted online July 15, 2021. Accessed August 10, 2022. https://dx.doi.org/10.2139/ssrn.3887443
- 47↑
Le Boedec K. Reference interval estimation of small sample sizes: a methodologic comparison using a computer-simulation study. Vet Clin Pathol. 2019;48(2):335–346.
- 48↑
Fagerland MW. t-tests, non-parametric tests, and large studies—a paradox of statistical practice? BMC Med Res Methodol. 2012;12:78.