Published online Jan 02, 2024.
https://doi.org/10.3348/kjr.2023.0911
Prediction of Venous Trans-Stenotic Pressure Gradient Using Shape Features Derived From Magnetic Resonance Venography in Idiopathic Intracranial Hypertension Patients
Abstract
Objective
Idiopathic intracranial hypertension (IIH) is a condition of unknown etiology associated with venous sinus stenosis. This study aimed to develop a magnetic resonance venography (MRV)-based radiomics model for predicting a high trans-stenotic pressure gradient (TPG) in IIH patients diagnosed with venous sinus stenosis.
Materials and Methods
This retrospective study included 105 IIH patients (median age [interquartile range], 35 years [27–42 years]; female:male, 82:23) who underwent MRV and catheter venography complemented by venous manometry. Contrast enhanced-MRV was conducted under 1.5 Tesla system, and the images were reconstructed using a standard algorithm. Shape features were derived from MRV images via the PyRadiomics package and selected by utilizing the least absolute shrinkage and selection operator (LASSO) method. A radiomics score for predicting high TPG (≥ 8 mmHg) in IIH patients was formulated using multivariable logistic regression; its discrimination performance was assessed using the area under the receiver operating characteristic curve (AUROC). A nomogram was constructed by incorporating the radiomics scores and clinical features.
Results
Data from 105 patients were randomly divided into two distinct datasets for model training (n = 73; 50 and 23 with and without high TPG, respectively) and testing (n = 32; 22 and 10 with and without high TPG, respectively). Three informative shape features were identified in the training datasets: least axis length, sphericity, and maximum three-dimensional diameter. The radiomics score for predicting high TPG in IIH patients demonstrated an AUROC of 0.906 (95% confidence interval, 0.836–0.976) in the training dataset and 0.877 (95% confidence interval, 0.755–0.999) in the test dataset. The nomogram showed good calibration.
Conclusion
Our study presents the feasibility of a novel model for predicting high TPG in IIH patients using radiomics analysis of noninvasive MRV-based shape features. This information may aid clinicians in identifying patients who may benefit from stenting.
INTRODUCTION
Idiopathic intracranial hypertension (IIH) is a condition marked by an elevated pressure of the cerebrospinal fluid (CSF), without any apparent structural etiology [1]. This condition may be manifested by symptoms such as pulsatile tinnitus, persistent headache, and visual disturbances [2, 3]. Although the exact pathophysiologic mechanism underlying IIH remains unclear, venous stenosis is considered a potential contributor [4, 5]. Recent findings have validated the effectiveness of stenting as a therapeutic approach for patients diagnosed with IIH who also exhibit concurrent venous sinus stenosis [6]. Before stenting, the trans-stenotic pressure gradient (TPG) must be measured to quantify the severity of stenosis and determine treatment suitability [7]. Venous sinus stenting is typically considered for patients exhibiting a TPG exceeding 8 mmHg, referred to as a high TPG [8].
Pressure measurements determined using digital subtraction angiography (DSA) are required to accurately assess TPG. However, it is difficult to use in the clinical practice because of its invasive nature. Despite recent attempts to identify alternatives to DSA for measuring the TPG [9, 10, 11], studies that quantitatively evaluate the degree of stenosis remain limited. Consequently, the investigation of non-invasive TPG prediction methods is of significant clinical interest.
This study aimed to quantitatively assess venous stenosis using shape features obtained from magnetic resonance venography (MRV) images to examine the association between these features and high TPG and establish a radiomics score for predicting high TPG in IIH patients. Subsequently, we aimed to devise a practical nomogram tailored for clinical applications by integrating clinical characteristics with the radiomics scores.
MATERIALS AND METHODS
Patients
This study was approved by the ethical committee of Beijing Tiantan Hospital, Capital Medical University (ethical statement number: KY2016-039-02). Conducted in alignment with the ethical principles outlined in the 1964 Helsinki Declaration, this research obtained written informed consent from all participants. Patients from a preexisting registry were retrospectively selected [12]. All patients in the registry were diagnosed with IIH based on a comprehensive set of criteria: 1) papilledema; 2) normal neurological examination, except for cranial nerve abnormalities; 3) neuroimaging revealing normal brain parenchyma without hydrocephalus, mass, or any structural lesion; 4) normal CSF composition; 5) elevated CSF opening pressure (greater than 250 mmH2O); and 6) additional neuroimaging features supporting an IIH diagnosis. In addition, a diagnosis of IIH can be made even in the absence of papilledema if criteria 2 and 3 are satisfied for unilateral or bilateral abducens nerve palsy. In this study, patients in the registry who met the following inclusion criteria were selected: 1) patients with preprocedural MRV, and 2) patients who underwent catheter venography accompanied by venous manometry. The exclusion criteria were as follows: 1) poor MRV quality and 2) missing baseline information. A flowchart of the patient inclusion process is shown in Figure 1.
Fig. 1
Flowchart of participants in this study. This flowchart details the progression and categorization of participants throughout the study. IIH = idiopathic intracranial hypertension, MRV = magnetic resonance venography
Patients meeting the inclusion criteria were bifurcated into two distinct groups, defined by a TPG cutoff value of 8 mmHg—a commonly used indicator for stenting [13]. In particular, the first group included patients with a TPG lower than 8 mmHg, while the latter encompassed those presenting with a TPG of 8 mmHg or higher. Furthermore, random allocation was employed to distribute these eligible patients into the training and test datasets, maintaining a ratio of 7:3.
MRV Examination and Feature Extraction
Contrast enhanced-MRV was conducted under 1.5 Tesla system (Siemens), and the images were reconstructed using a standard algorithm. The scanning parameters included a 432 × 432 matrix, slice thickness of 1.6 mm, echo time of 1 ms, repetition time of 2.3 ms, and flip angle of 25°. A maximum dose of 20 mL (0.2 mmol/kg) of the gadolinium contrast agent (dimeglumine gadopentetate injection, Beilu Pharmaceutical) was administered using an autoinjector at a rate of 2 mL/s. This was immediately followed by a 20 mL saline flush, employing the same injection parameters. Images were captured in the sagittal plane pre- and post-contrast administration and subsequently reconstructed using a standard algorithm.
The stenotic regions were manually segmented by neuroradiologists who were blinded to the pressure measurement using 3D Slicer (version 4.10.2; http://www.slicer.org). For each patient, three-dimensional (3D) shape features were obtained using the Python-based PyRadiomics package (version 3.0; https://github.com/Radiomics/pyradiomics).
Venous Manometry
Manometry was performed under local anesthesia in all patients. Typically, a Rebar-27 microcatheter (Medtronic) was deftly navigated over a microwire to enable access to the venous sinus. The central part of Rebar-27 (Medtronic) was interfaced with a DELTRAN II pressure-monitoring kit (Utah Medical Products). Subsequently, a contrast medium was infused through the microcatheter to facilitate venographic assessment of the venous sinus.
Construction of Models for Predicting High TPG
To reduce the number of features, the three-fold cross-validation least absolute shrinkage and selection operator (LASSO) algorithm—a variant of multivariable linear regression—was employed in the training dataset. The lambda value that produced the highest cross-validated discrimination performance in the LASSO fit was chosen, and features retaining nonzero coefficients were further utilized for model construction. Following multivariable analysis, features with a P-value of < 0.05 were incorporated into the model construction. The radiomics score was then determined based on the acquired coefficients:
Subsequently, the Youden’s Index was utilized to establish the optimal threshold for the radiomics score. This threshold was determined as the radiomics score that resulted in the maximum Youden’s J-statistic (J). The calculated cut-off point was subsequently utilized for the test dataset, and the performance of the radiomics score was evaluated. Additionally, metrics derived from the confusion matrix were calculated.
We developed a nomogram for predicting the risk of high TPG levels in IIH patients based on a multivariable logistic regression model. The nomogram incorporated the radiomics score and significant clinical predictors (P < 0.05) from univariable analysis. This model assigns a coefficient to each predictor that quantifies its influence on the outcomes. A positive coefficient indicates an increased likelihood of an outcome, whereas a negative coefficient suggests the opposite. Coefficients from the logistic regression model were then used to construct a nomogram. The most influential predictor (with the largest absolute coefficient) set the scale and was assigned a maximum score of 100 points. The other predictors were scaled relative to the maximum based on their respective coefficients. To utilize the nomogram, a vertical line was drawn from each predictor to the corresponding score on the “Points” scale. The total score was obtained by summing the scores for each predictor, which was used to estimate the probability of high TPG levels in IIH patients. A comprehensive depiction of the methodology used in this study is presented in Figure 2.
Fig. 2
Diagram of this study workflow comprising several stages. Initially, venous sinuses are reconstructed utilizing magnetic resonance venography, and regions of interest (ROIs) within the sinuses are delineated. Next, three-dimensional shape features are extracted from the identified ROIs, and the least absolute shrinkage and selection operator method is employed to select informative features for constructing a radiomics score. Finally, a clinically applicable nomogram is established by integrating clinical characteristics with the radiomics score. AUC = area under the curve
Statistical Analyses
All statistical analyses were performed using R software (version 4.0.1; R Foundation for Statistical Computing), with the implementation of relevant packages including “glm,” “OptimalCutpoints,” and “ggplot2.” Differences in continuous variables were assessed using either Student’s t-test or the Mann–Whitney U test, as appropriate. Continuous data, based on their distribution, are presented as means with standard deviations or medians with interquartile ranges (IQRs). Categorical variables were compared using the chi-square test, and the results are reported as event counts with corresponding relative frequencies (%).
The discrimination performances of the radiomics score and nomogram were assessed using the area under the receiver operating characteristic curve (AUROC). Regarding the calibration performance of the nomogram, we used a calibration curve to assess how close the predicted probabilities were to the actual outcomes. We generated this using the “calibrate” function from the “rms” package, and the results were plotted to visually assess the nomogram's calibration. The closer the calibration curve is to a 45-degree diagonal line, the better the model calibration. Bootstrap validation with 1000 resamples was performed to assess the reliability and stability of the nomogram. A P-value of < 0.05 was set as the threshold for statistical significance.
RESULTS
Baseline Characteristics and Clinical Features
A total of 105 patients were included in the study, of whom 82 (78.1%) were female. The median age at presentation was 35 years (IQR: 27–42 years). Of the total population, 36 (34.3%) had hypertension, 2 (1.9%) had diabetes mellitus, 9 (8.6%) had hyperlipidemia, and 3 (2.9%) had rheumatological autoimmune diseases. Patients exhibited a median body mass index of 29.0 (IQR, 25.3–30.5) kg/m2. At initial presentation, headache, cervicodynia, tinnitus, papilledema, and vision impairment were reported in 31 (29.5%), 21 (20.0%), 15 (14.3%), 13 (12.4%), and 92 (87.6%) patients, respectively. Twenty-five (23.8%) patients were using anticoagulant medications. Our study found that stenosis of the right transverse sinus was more prevalent than left-sided stenosis, with the junction of the transverse and sigmoid sinuses being the most frequent location of stenosis. Of the total population, 72 (68.6%) patients had a TPG of ≥ 8 mmHg and 33 (31.4%) had a TPG of < 8 mmHg.
Of the 105 patients, 73 (70%; 50 and 23 with and without high TPG, respectively) were randomly allocated to the training dataset, and the remaining 32 (30%; 22 and 10 with and without high TPG, respectively) constituted the test dataset. The baseline characteristics of both datasets were comparable, with no statistically significant disparities noted. For all 105 patients, the univariable analysis demonstrated a significant difference between sex and TPG (P = 0.018), in which female sex was associated with a higher-pressure gradient (odds ratio [OR] = 3.17; 95% confidence interval [CI], 1.22–8.25). Regarding clinical presentation, cervicodynia (P = 0.001; OR = 12.31; 95% CI, 1.58–96.17) was associated with an increased TPG. Anticoagulant use (P = 0.046; OR = 2.99; 95% CI, 0.93–9.55) also exhibited an association with a higher TPG. Table 1 presents the findings derived from the univariable analysis.
Table 1
Baseline characteristics and clinical features of the patients
Radiomics Score
The LASSO algorithm selected three informative features from the 14 shape features: least axis length, sphericity, and maximum 3D diameter. Figure 3 illustrates the distribution of the 14 shape features in the training dataset for cases with intracranial pressures (TPG) of ≥ 8 mmHg and < 8 mmHg (Fig. 3A) as well as the feature selection process using the LASSO algorithm for these 14 features (Fig. 3B, C). Figure 4 shows a schematic representation of the specific meanings of the three selected shape features. Within the training dataset, multivariable logistic regression was performed utilizing these three characteristics to ascertain the coefficients and intercepts. Following this, the radiomics scores were calculated as follows:
Fig. 3
Analysis of shape features and trans-stenotic pressure gradient in the training dataset. A: Distribution of the 14 shape features in the training dataset for cases with a trans-stenotic pressure gradient (TPG) of ≥ 8 and those with a TPG of < 8. B: The curve is plotted against log (λ), with three features selected based on non-zero coefficients using the minimum criteria. C: Feature profiles from the least absolute shrinkage and selection operator are shown, where each colored line denotes an individual feature's coefficient. *P < 0.05, **P < 0.01, ***P < 0.001. AUC = area under the curve
Fig. 4
Illustration of 3D-shape features. Sphericity is a measure of the roundness of the shape of the 3D-ROI relative to a sphere. It is a dimensionless measure, independent of scale and orientation. The value range is 0 to 1, where a value of 1 indicates a perfect sphere. Maximum 3D diameter is defined as the largest pairwise Euclidean distance between ROI surface mesh vertices. Least axis length yields the smallest axis length of the ROI-enclosing ellipsoid and is calculated using the largest principal component. ROI = region of interest
In the training dataset, the optimal cut-off point for the radiomics score was 0.281. Using this cutoff point, the AUROC was 0.906 (95% CI, 0.836–0.976). We applied the same cutoff point to the test dataset, resulting in an AUROC of 0.877 (95% CI, 0.755–0.999). The AUROC of the radiomics scores in the training and test datasets and bar plots for both datasets illustrating the prediction values for each patient are presented in Figure 5.
Fig. 5
Radiomics score prediction and AUROC curves in the training and test datasets. Bar plots for the training (A) and test datasets (B) with the prediction value of the radiomics score in each patient. The receiver operating characteristic curves of the radiomics score in the training (C) and test datasets (D). TPG = trans-stenotic pressure gradient, ROC = receiver operating characteristic curve, TPR = true positive rate, AUROC = area under the receiver operating characteristic curve, CI = confidence interval, FPR = false positive rate
Nomogram Incorporating the Radiomics Score and Clinical Features
We developed a nomogram for predicting the risk of high TPG levels in IIH patients based on a multivariable logistic regression model. The nomogram integrated the following variables: sex, cervicodynia, anticoagulant use, and radiomics score. The respective coefficients were as follows: intercept (-1.1871), female (1.2285), cervicodynia (2.0759), anticoagulant usage (0.5788), and radiomics score (0.9721). Bootstrap validation using 1000 resamples indicated a corrected AUROC of 0.917 for the training dataset and 0.901 for the test set. Figure 6 shows the nomogram and its corresponding calibration curve.
Fig. 6
Nomogram and its calibration curve. A: The nomogram incorporates four entries: sex, cervicodynia, anticoagulant use, and radiomics score. To utilize the nomogram, start with the specific value of each predictor, draw a line straight up to the “Points” row to determine the score for that predictor. Sum the scores from all predictors to obtain a total score. Finally, draw a line from the “Total Points” row straight down to find the corresponding predicted probability of high TPG risk. B, C: The calibration curves depicts the relationship between the predicted risk of high TPG from the nomogram and the actual observed risk. A 45-degree line represents perfect calibration, where the model's predicted probabilities exactly match the actual observed probabilities. TPG = trans-stenotic pressure gradient
DISCUSSION
In this study, we developed models for predicting high TPG levels in IIH patients using shape features extracted from non-invasive MRV data. The predictive model demonstrated good performance on the test set. Furthermore, we established a nomogram incorporating the shape-based radiomics score and three clinical features to assist in clinical decision making.
Although stenting has become a promising therapeutic avenue for patients diagnosed with venous sinus stenosis, there is a certain degree of uncertainty regarding the guidelines for patient selection. West et al. [14] analyzed a prospectively collected database and discovered that each 10% increase in stenosis was associated with a rise of 3.5 mmHg in the TPG. Zhao et al. [15] found that longer transverse sinus stenosis (TSS) resulted in a larger TPG, possibly owing to the increased resistance presented by the narrowed transverse sinus. The TPG is a crucial indicator for assessing the severity of venous sinus stenosis, provides valuable guidance for clinical decision-making, and is important in the field of neurointerventional surgery. As a result, venous sinus manometry is commonly conducted to evaluate eligibility for treatment, with a TPG ≥ 8 mmHg serving as a criterion for selecting candidates for stenting [6].
Although DSA venous manometry remains the gold standard for measuring TPG, its invasive nature can lead to complications, limiting its extensive application in clinical settings. Although computed tomography venography and MRV provide valuable insights into the anatomy and morphology of the cerebral venous sinuses, enabling the evaluation of stenosis and its potential hemodynamic implications, most current research on venous sinus stenosis predominantly focuses on subjective qualitative imaging assessments [16, 17, 18]. Ding et al. [7] presented a novel approach for evaluating the TSS using computed tomography venography, with a focus on TPG. They utilized the post-stenotic segment of the transverse sinus as a reference point and discovered a robust correlation between its cross-sectional area and TPG. Several recent studies have attempted to develop quantitative MRV-based methods to evaluate the severity of venous sinus stenosis and estimate TPG values. In a study investigating the hemodynamic characteristics of the TSS, researchers employed four-dimensional flow magnetic resonance imaging (MRI) of the venous sinus [19]. They identified a correlation between the blood flow velocity and TPG. Their findings suggested that four-dimensional flow MRI might be an appropriate tool for determining which patients with TSS could benefit from invasive venous manometry.
Numerous studies have highlighted the significance of shape features in relation to cerebrovascular disease [20, 21]. Ou et al. [20] illustrated how shape features derived from computed tomography angiography could adeptly predict aneurysm stability. Similarly, Liu et al. [21] successfully extracted morphological features using PyRadiomics, discovering that these features were similarly predictive in assessing aneurysm stability. In this study, we quantitatively evaluated venous sinus stenosis using shape features derived from MRV. Three 3D shape features—sphericity, maximum 3D diameter, and least-axis length—were selected to develop a radiomics score for predicting TPG in IIH patients. Sphericity, a dimensionless measure of an object’s roundness relative to a perfect sphere, was found to have an inverse relationship with the TPG. Higher sphericity suggests that the shape of the stenosed transverse sinus more closely resembles a sphere. The spherical shape of the venous sinus may provide a more uniform cross-sectional area, minimizing the flow resistance within the stenosed region [22]. Lower flow resistance, therefore, promotes blood flow and potentially reduces the pressure gradient across a narrow area. Similarly, the maximum 3D diameter inversely correlated with TPG. A larger maximum 3D diameter implies a more spacious cross-sectional area for blood flow, thereby reducing the resistance and pressure gradient. Consequently, blood can flow more efficiently, resulting in a lower overall resistance and TPG [19]. The least axis length is another key predictor. A larger least axis length indicates a more favorable geometry for blood flow in the stenosed region, allowing for lower resistance and TPG [23]. Decreasing the likelihood of turbulent flow and pressure buildup may contribute to better TPG management. These findings offer valuable insights into the pathophysiological processes underlying sinus stenosis and warrant further investigation for potential clinical applications.
Considering the potential relationship between the clinical characteristics and TPG in IIH patients, our study also evaluated the association between TPG and various clinical features. We developed an easy-to-use nomogram for clinical applications by combining radiomics scores with relevant clinical characteristics. In this study, we identified the following three clinical features associated with TPG in IIH patients: sex, anticoagulant use, and cervicodynia. Female patients exhibited a higher risk of elevated TPG levels, potentially owing to hormonal factors such as estrogen affecting vascular tone and cerebral venous sinus thrombosis [24]. Hormonal fluctuations during the menstrual cycle might also influence intracranial pressure and blood flow, impacting TPG. Although not widely reported, our study identified a link between anticoagulant use and elevated TPG among IIH patients exhibiting venous sinus stenosis. A possible explanation for this association is that anticoagulants alter the hemodynamics within the region of venous sinus stenosis, subsequently influencing TPG [25, 26]. An increased TPG may lead to venous congestion and engorgement of the cervical venous plexus, causing neck pain. Venous sinus stenosis can also alter CSF dynamics, resulting in cervicodynia due to changes in intracranial and spinal pressures [27]. By merging clinical and radiomics data, our model offers a more holistic view of patient information, enhancing its stability and broadening its applicability. The incorporation of various features bolsters the robustness of the model against potential variability. The resulting nomogram is a dependable tool for clinicians and demonstrates excellent discrimination, calibration, and reliability through bootstrap validation.
Our study had some limitations. First, the study was limited by its relatively small sample size and retrospective nature. Second, the study lacked external validation. Third, the accuracy and reliability of non-invasive imaging methods for predicting TPG across different patient populations and stenosis subtypes require further validation. Future prospective multicenter studies are warranted to address these limitations and further refine the predictive model.
In conclusion, this study presented the feasibility of a novel model for predicting high TPG in IIH patients using radiomics analysis of non-invasive MRV-derived shape features. Additionally, we developed a nomogram incorporating clinical factors and radiomics scores based on three informative shape characteristics. This approach could provide valuable insights for clinicians in selecting patients for venous sinus stenting, possibly reducing their reliance on invasive procedures. Future studies should include larger sample sizes and more diverse patient populations to validate and optimize the proposed model.
Conflicts of Interest:The authors have no potential conflicts of interest to disclose.
Author Contributions:
Conceptualization: Chao Ma.
Data curation: Haoyu Zhu.
Formal analysis: Chao Ma.
Funding acquisition: Yupeng Zhang.
Investigation: Yuzhou Chang.
Methodology: Chuhan Jiang.
Project administration: Shikai Liang.
Software: Yupeng Zhang.
Supervision: Dapeng Mo.
Validation: Yupeng Zhang.
Writing—original draft: Haoyu Zhu.
Writing—review & editing: Chao Ma.
Funding Statement:This work was supported by the National Natural Science Foundation of China (NSFC) Youth Project (grant number 82301457).
Availability of Data and Material
The datasets generated or analyzed during the study are available from the corresponding author on reasonable request.
References
-
Bussière M, Falero R, Nicolle D, Proulx A, Patel V, Pelz D. Unilateral transverse sinus stenting of patients with idiopathic intracranial hypertension. AJNR Am J Neuroradiol 2010;31:645–650.
-
-
Ding H, Zhao P, Lv H, Li X, Qiu X, Zeng R, et al. Correlation between trans-stenotic blood flow velocity differences and the cerebral venous pressure gradient in transverse sinus stenosis: a prospective 4-dimensional flow magnetic resonance imaging study. Neurosurgery 2021;89:549–556.
-
-
Eisenman DJ, Raghavan P, Hertzano R, Morales R. Evaluation and treatment of pulsatile tinnitus associated with sigmoid sinus wall anomalies. Laryngoscope 2018;128 Suppl 2:S1–S13.
-
-
Sun SS, Liu FY, Tsai JJ, Yen RF, Kao CH, Huang WS. Using 99mTc HMPAO brain SPECT to evaluate the effects of anticoagulant therapy on regional cerebral blood flow in primary antiphospholipid antibody syndrome patients with brain involvement-a preliminary report. Rheumatol Int 2003;23:301–304.
-
-
Bateman GA. Vascular hydraulics associated with idiopathic and secondary intracranial hypertension. AJNR Am J Neuroradiol 2002;23:1180–1186.
-