FormalPara Key Summary Points

Why carry out this study?

There remains a critical need for precise localization of the epileptogenic zone in individuals with drug-resistant epilepsy (DRE).

Visual-based qualitative analysis is time-consuming and strongly influenced by physician experience. Therefore, we aimed to evaluate the efficacy of CortexID quantitative analysis in the localization of the epileptogenic zone in patients with TLE.

What was learned from the study?

CortexID quantitative analysis achieves diagnostic efficacy comparable to visual analysis in the localization of the epileptogenic zone in patients with TLE.

CortexID quantitative analysis can provide additional information in detecting the epileptogenic zone.

Nuclear medicine physicians can choose to supplement their reading methodology with the CortexID quantitative analysis.

Introduction

Epilepsy is a common brain disorder that results in persistent seizures, impacting patients’ quality of life [1, 2]. While anti-seizure medication can control seizures in many cases, about one-third of patients continue to experience uncontrolled seizures. For individuals with drug-resistant focal epilepsy, epilepsy surgery offers the best chance of long-term freedom from seizures. Successful surgery relies on accurately identifying the epileptogenic zone and completely removing or destroying it [1, 3,4,5].

Epileptogenesis is often marked by neuronal damage and gliosis [6], which can result in reduced brain metabolism. 18F-Fluorodeoxyglucose positron emission tomography (FDG-PET) imaging can reveal hypometabolic regions during the interval between seizures in patients with epilepsy, and a coincidence rate of up to 90% between PET localization of the epileptogenic zone and intraoperative cortical electroencephalography (EEG) can be achieved [7]. Visual-based qualitative analysis is the most widely used method for PET analysis at present. However, this approach is time-consuming, and is strongly influenced by physician experience.

Quantitative analysis with PET imaging is a promising approach for evaluating the epileptogenic zone. Visual assessment of hypometabolism using 18F-FDG PET is less accurate than quantitative assessment using voxels, machine learning, model classifiers, statistical parametric mapping (SPM), and other artificial intelligence methods [8,9,10,11]. However, these methods are cumbersome and time-consuming, and some of them need to build a normal PET image population database separately, so they are not easy to apply promptly in the clinic [12].

CortexID Suite is a quantitative analysis software developed by GE Healthcare (Chicago, IL, USA) that enables fast data processing without the need to develop or purchase additional software. The software supports information on the relative changes in the metabolism of brain FDG and amyloid images and helps to evaluate PET imaging of the human brain [13]. The software has proprietary, adaptive template alignment technology based on voxel or three-dimensional stereotactic surface projection (3D-SSP) to normalize PET images by anatomy to a standardized space [14]. In a recent study, Wang et al. [15] compared epileptic focus detection by SPM and 3D-SSP for 18F-FDG-PET brain mapping analyses and concluded that both methods can improve the detection rate, recommending the use of 3D-SPP and SPM for more complex cases. CortexID quantitative analysis software has been applied to a variety of neurological diseases, including ischemic-hypoxic encephalopathy and Alzheimer’s disease [13, 14, 16, 17]. However, there is still a lack of studies on its application in locating the epileptogenic zone.

Considering the multifactorial nature of the disease, integrating multiple levels of information is crucial for localizing the epileptogenic zone. Therefore, the aim of this study was to evaluate the efficacy of CortexID quantitative analysis in the localization of the epileptogenic zone in patients with temporal lobe epilepsy (TLE).

Materials and Methods

Data Selection and Patients

Data on patients with epilepsy who underwent 18F-FDG PET examination from August 2021 to September 2023 at the Department of Nuclear Medicine, Xijing Hospital, were retrospectively collected. This study was approved by the Medical Ethical Committee of the First Affiliated Hospital of Air Force Military Medical University. The inclusion criteria for this study were as follows: (1) meeting the diagnostic criteria for epilepsy [18]; (2) 18F-FDG PET examination performed in the interictal period; (3) clearly identified as a patient with focal epilepsy; (4) clear localization of epileptic foci; and (5) having signed an informed consent form. Exclusion criteria were as follows: (1) patients with surgical resection or stereoelectroencephalography radio-frequency thermocoagulation; (2) patients with other serious neurological impairments or neurological diseases; (3) incomplete collection of clinical data.

Location of the Epileptogenic Zone

Clinical information, including age, duration of disease, age of onset, seizure duration, seizure frequency, medication regimen, history, 24-h video EEG, brain nuclear magnetic resonance imaging (MRI), and seizure semiology, was collected from the Xijing Hospital medical record system. The assumed epileptogenic zone was evaluated comprehensively by two skilled neurologists in preoperative assessment of epilepsy based on these findings, and in the case of disagreement, a third expert was invited for discussion.

18F-FDG PET Acquisition

Image acquisition of patients was conducted using PET/MR scanners. All patients had fasted for more than 6 h, fasting blood glucose was controlled below 11.1 mmol/L, and 18F-FDG was injected intravenously. After the injection, the scan was started after 40–60 min of rest with eyes closed in a suitable temperature environment. All patients signed informed consent before the examination.

Image Analysis

Visual Analysis Method

Two experienced nuclear medicine physicians interpreted 18F-FDG PET images indicating whether the patient had a hypometabolic zone in the temporal lobe; in cases of disagreement, a third nuclear medicine physician was required to make a decision.

CortexID Quantitative Analysis

CortexID Suite quantitative analysis is a cerebral metabolic analysis software that analyzes 18F-FDG PET brain micrographs using the 3D-SSP method [19], with an adaptive template alignment technique by anatomical normalization to a standardized space. The Z-score method compares the patient’s standardized cortical peak metabolic activity with that of a healthy control group from the software’s normal group database, displaying the hypometabolic Z-score maps at eight standard angles [20, 21]. All of the above processes were automatically processed and displayed by CortexID (Fig. S1 in the Supplemental Material). In this study, the difference in the Z-score between the left and right temporal lobes was used to determine a quantitative analysis result.

Calculation and Statistical Analysis

Statistical analysis was conducted using SPSS 26.0 software and R programming language version 4.3.1. Continuous variables were first tested for conformity with normal distribution and, if not normally distributed, were described as median (interquartile spacing) [M(P25, P75)]; conformity with normal distribution was described as mean ± standard deviation (x ± SD) and compared using Student’s t-test. Otherwise, the Mann–Whitney U test was used. Categorical variables were expressed as proportions. Comparisons between the two groups were performed using the X2 or Fisher exact test, as appropriate. The area under the receiver operating characteristic (ROC) curve (AUC) area was compared using the DeLong test, and all tests were two-tailed, with p < 0.05 indicating a statistically significant difference.

Results

Construction of Diagnostic Models for TLE

Calculation of Thresholds for Quantitative Analyses of Medial Temporal Lobe Epilepsy (MTLE)

A total of 102 patients with epilepsy were included, of which nine had bilateral MTLE lesions. Therefore, a total of 93 cases were included in the diagnostic model, including 40 patients with MTLE. The ROC curves were constructed according to the assumed epileptogenic zone (Fig. 1): AUC = 0.905, Z-score threshold = 0.87 for the MTLE. The Z-score difference between the left and right MTLE was used to determine whether there were lesions in the medial temporal lobe, with the plus and minus signs representing the lateralization. An absolute value greater than the threshold value was considered as indicating the presence of lesions in the medial temporal lobe.

Fig. 1
figure 1

ROC curve for quantitative analysis of CortexID in the MTLE. AUC = 0.905 for the MTLE. ROC Receiver operating characteristic; MTLE Medial Temporal Lobe Epilepsy; AUC Area under the curve

Full size image

Construction of the MTLE Model

Univariate analysis showed statistically significant (p < 0.05) results for MRI, PET visual analysis, and CortexID quantitative analysis (Table S1 in the Supplemental Material). The multiple linear regression analysis showed that the variance inflation factor (VIF) of each variable was less than 5, indicating no collinearity issue (Table S2 in the Supplemental Material). Therefore, three factors were included in the binary logistic regression model for multifactorial analysis, indicating that the results of the PET visual analysis and CortexID quantitative analysis had an independent diagnostic basis (p < 0.05) (Table S3 in the Supplemental Material).

In this study we chose to use logistic regression and cross-validation to construct the model and calculate AUC. The results are shown in Table S4 in the Supplemental Material, indicating that the logistic regression results are superior. Therefore, logistic regression was used to construct the ROC curve, as shown in Fig. 2, and the AUC was calculated. The accuracy of the MTLE visual analysis was compared with the accuracy of each diagnostic model as shown in Table 1. Statistical results showed the following: AUC (visual analysis) = 0.922 and AUC (CortexID quantitative analysis) = 0.853, with no statistically significant difference (p = 0.136); AUC (visual analysis) = 0.922 and AUC (combination of both) = 0.971, with a statistically significant difference (p = 0.013).

Fig. 2
figure 2

ROC curve for different diagnostic indicators of medial temporal lobe epilepsy. AUC (visual analysis) = 0.922, AUC (CortexID quantitative analysis) = 0.853, AUC (combination of both) = 0.971. ROC Receiver operating characteristic; AUC Area under the curve

Full size image
Table 1 Comparison of the AUC of different diagnostic indicators in the MTLE with the AUC of PET visual analysis
Full size table

Calculation of Thresholds for Quantitative Analyses of Lateral Temporal Lobe Epilepsy (LTLE)

A total of 102 patients with epilepsy were included, of which seven had bilateral LTLE lesions. Therefore, a total of 95 cases were included in the diagnostic model, among which 47 patients had LTLE. The ROC curves were constructed according to the assumed epileptogenic zone (Fig. S2 in the Supplemental Material): AUC = 0.843, Z-score threshold = 2.175 for the LTLE.

Construction of the LTLE Model

Univariate analysis showed statistically significant (p < 0.05) results for seizure duration of 1–3 min, MRI, PET visual analysis, and CortexID quantitative analysis (Table S5 in the Supplemental Material).

The multiple linear regression analysis showed that the VIF of each variable was less than 5, indicating no collinearity issue (Table S6 in the Supplemental Material). Therefore, four factors were included in the binary logistic regression model for multifactorial analysis, indicating that the results of the PET visual analysis and CortexID quantitative analysis had an independent diagnostic basis (p < 0.05) (Table S7 in the Supplemental Material).

In this study we chose to use logistic regression and cross-validation to construct the model and calculate AUC. The results are shown in Table S4 (in the Supplemental Material), indicating that the logistic regression results are superior. Therefore, logistic regression was used to construct the ROC curve, as shown in Fig. S3 in the Supplemental Material, and the AUC was calculated. The accuracy of the LTLE visual analysis was compared with each diagnostic model as shown in Table 2. Statistical results showed the following: AUC (visual analysis) = 0.842 and AUC (cortexID quantitative analysis) = 0.831, with no statistically significant difference (p = 0.808); AUC (visual analysis) = 0.842 and AUC (combination of both) = 0.897, with a statistically significant difference (p = 0.009).

Table 2 Comparison of the AUC of different diagnostic modalities in the LTLE with the AUC of visual analysis by PET
Full size table

Surgical Patients

A total of 23 patients who underwent surgery were excluded from this study, which was used for the validation part of the above study. There were 29 lesions in the 23 patients (detailed clinical information can be found in Table S8 in the Supplemental Material), including six lesions outside the temporal lobe and 23 lesions in the temporal lobe (11 lesions in the medial temporal lobe and 12 lesions in the lateral temporal lobe). After follow-up (12 [6, 15] months), all of these patients had a good outcome with regard to their seizures: 21 patients with Engel I, two patients with Engel II, and no patients with Engel III–IV. Therefore, the reliability of preoperative evaluation for this group of patients is acceptable. Statistical results showed AUC (visual analysis) = 0.822 and AUC (cortexID quantitative analysis) = 0.75, and the difference was not statistically significant (p = 0.377). The sensitivity of CortexID quantitative analysis reached 100%, which was higher than that of visual analysis in Table 3 and Table S9 (in the Supplemental Material). A lesion with negative visual analysis and positive quantitative analysis is shown in Figs. 3 and 4. This patient’s PET visual analysis suggests only hypometabolic regions in the left medial temporal lobe, and CortexID quantitative analysis shows the presence of lesions in both the left medial and lateral temporal lobes. The pathological results indicated a low-grade glial neuronal tumor in the medial left temporal lobe and focal cortical dysplasia in the lateral left temporal lobe. No seizures were observed in the patient after 12 months of postoperative follow-up, and the prognosis was Engel I.

Table 3 Comparison of diagnostic efficacy of PET visual analysis and CortexID quantitative analysis of the TLE
Full size table
Fig. 3
figure 3

A 24-year-old patient with refractory focal epilepsy for 2 years. PET visual analysis of the patient found hypometabolic regions only in the left medial temporal lobe (a, b). PET positron emission tomography

Full size image
Fig.4
figure 4figure 4

A 24-year-old patient with refractory focal epilepsy for 2 years. CortexID quantitative analysis showed a difference of 6.19 in the Z-score between the left and right sides of the medial temporal lobe of the patient, while a Z-score difference of 2.76 was found between the left and right sides of the lateral temporal lobe, indicating the presence of lesions in both the left medial and lateral temporal lobes

Full size image

Ethical Approval

We confirm that we have read the Journal’s position on issues involved in ethical publication and affirm that this report is consistent with those guidelines. This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Medical Ethical Committee of the First Affiliated Hospital of the Air Force Military Medical University (reference no. KY20222137).

Discussion

In this study, CortexID quantitative analysis achieved diagnostic efficacy comparable to visual analysis in the localization of the epileptogenic zone in patients with TLE. The results of postoperative pathology and follow-up suggest that CortexID quantitative analysis can provide additional information in epileptogenic zone detection.

CortexID quantitative analysis was as effective as visual analysis in diagnosing TLE in this study, suggesting that CortexID quantitative analysis is a non-inferior diagnostic modality. The results of the visual analysis are closely related to the experience of the doctor who reads the image and are highly influenced by the doctor’s subjectivity. In contrast, CortexID quantitative analysis is not affected by the subjective factors of the doctor. It is a repeatable, time-saving, and simple-to-use method. Quantitative analysis is valuable in that it can interpret images in the case of low confidence in visual analysis. Thus, it is useful in improving the speed of image reading and identifying difficult cases. Moreover, our study indicates that combining CortexID quantitative analysis with visual analysis significantly further improves the diagnostic accuracy of epileptogenic zone localization.

In follow-up data from postoperative patients, we found that quantitative analysis is more sensitive in locating the epileptogenic zone than visual analysis, consistent with previous studies. Yen-Cheng et al. [10] used a machine learning classifier to extract image features from 18F-FDG PET data and to classify normalized PET uptake with regions of interest (ROIs) from 18F-FDG PET images to identify the lateralized epileptogenic zone. Using postoperative pathology as the gold standard, they found that quantitative analysis was more sensitive than visual analysis (98.15 vs. 81.48%, respectively). All patients in our study demonstrated good outcomes (Engel I–II) during follow-up. The sensitivity of quantitative analysis in the present study was as high as 100%, which may be related to the relatively small sample of postoperative patients and the risk of selection bias. In addition, the high sensitivity in postoperative patients may indicate that using postoperative follow-up as the gold standard is more reliable.

From a clinical point of view, nuclear medicine physicians can choose to supplement their reading methodology with the CortexID quantitative analysis tool to scrutinize images through visual means. In patients with epilepsy outside the temporal lobe, we can identify the site with a large metabolic difference between the left and right sides as a reduced metabolic area, while in bilateral lesions we compare the metabolic situation of the whole brain, focusing on the area with a large difference from the whole brain. The analysis of inconsistencies and the added value of considering visually negative and quantitatively positive cases as potentially more risky adds a tool to routine assessment. Compared to other studies [22], the greatest advantage of this study is that the quantitative analysis software used is convenient to use. The software already has a database of normal human metabolism, which can quickly draw preliminary conclusions for clinical doctors. There is no need for extensive modeling and image preprocessing, nor is there a need to rebuild a normal human database.

Several limitations of this trial should be pointed out. First, the control group of CortexID quantitative analysis software used in this study comes from a database for individuals aged 42–89 years with normal imaging [23]. This database was not accurately matched with the age and gender of the patients included in our study. However, the method used in this study was designed to calculate the metabolic differences between the left and right sides rather than simply comparing the differences between each site and the normal database. Thus, poor age matching in the control group may not have had a significant impact on the analysis [12]. Furthermore, a study on pediatric patients with epilepsy found that children over 6 years of age and adults appear to have the same pattern of glucose utilization [24]. Supplementing normal control data for additional age groups may further improve the accuracy of CortexID software.

Second, this is a retrospective study that considers the assumed epileptogenic zone as the gold standard. Although postoperative pathology and follow-up data may be more accurate in the localization of the epileptogenic zone, only a small proportion of our patients received surgery. Nonetheless, the results of quantitative analysis for the surgical and nonsurgical groups are similar.

Conclusion

In conclusion, the diagnostic efficacy of CortexID quantitative analysis is comparable to PET visual analysis in the localization of the epileptogenic zone in patients with TLE. CortexID analysis is repeatable, time-saving, and simple to use, and could be used as a non-inferior diagnostic modality to improve the accuracy of preoperative epileptogenic focus localization.