Content area
1. Introduction
The intracranial space has limited capacity, which means that any changes in the volume of its components—such as brain tissue, blood, or cerebrospinal fluid (CSF)—can cause serious health consequences [1,2]. In many neurological conditions, including brain tumors, strokes, neuroinflammatory disorders and traumatic brain injuries, a reduction in CSF reserve and the development of mass effect may occur, potentially resulting in severe complications [3,4]. Consequently, the assessment of CSF volume and the monitoring of its distribution are becoming critical elements of neuroimaging diagnostics.
In recent years, Artificial Intelligence (AI) has played an increasingly significant role in the analysis and interpretation of medical data. Through advanced algorithms, AI can automate the analysis of medical images and the identification of abnormalities in CSF volume and distribution [5,6,7]. Deep neural networks (DNNs) such as U-Net and its variants are the most commonly employed machine learning techniques in the segmentation of medical images and the analysis of CSF distribution [8]. These architectures are capable of accurately extracting anatomical structures due to their convolutional layers and ‘skip connections’, which help preserve spatial information [9]. In addition to convolutional networks, classical machine learning algorithms are also used—such as the random forest method, which classifies pixels or image regions based on a set of features including intensity, texture, and location, using an ensemble of decision trees [10]. As a result of its randomized feature selection and data sampling during training, this algorithm is highly resistant to overfitting and performs well on small datasets and tasks with limited computational demands [11]. The application of AI algorithms has the potential to accelerate the detection of subtle changes in CSF volume and identify potentially dangerous conditions, enabling more timely and effective clinical decision-making [12]. Automating these processes may not only improve diagnostic accuracy but also reduce response times, which can be crucial in urgent cases where timely intervention significantly affects treatment outcomes.
The use of AI for cerebrospinal fluid analysis on head CT represents a new and rapidly developing field of research. A synthesis of the current literature is therefore needed to outline the state of knowledge, summarize recent advances, and identify remaining challenges. In this review, we analyze studies employing AI for the segmentation and volumetric assessment of cerebrospinal fluid in head CT images, with the aim of evaluating whether these methods can support improvements in patient care during the diagnostic process of neurological diseases, as well as assess their potential benefits for radiology and diagnostic imaging professionals through process automation and workflow acceleration.
2. Materials and Methods
2.1. Search Strategy
In February 2025, a comprehensive search of medical databases was conducted, including MEDLINE via PubMed, Scopus, Web of Science, Embase, and the Cochrane Library.
The search strategy was developed in accordance with the PICO framework [13]. A detailed search strategy and the PICO scheme are provided (Supplement S1).
An initial total of 559 articles was retrieved across all databases. After applying filters, 307 articles remained. These were then imported into the Rayyan tool [14], which automatically identified 98 duplicates, of which 58 were removed.
Subsequently, 249 articles were screened by two independent reviewers. The screening process involved assessing titles, abstracts, and keywords. In total, 232 articles were excluded for the following reasons: wrong publication type (n = 18), inappropriate study design (n = 199), and paediatric population (n = 18).
Ultimately, 14 articles were included in the review (Supplement S2). All reviewer decisions were documented as follows: Reviewer 1—excluded: 224, included: 19, maybe: 6; Reviewer 2—excluded: 229, included: 6, maybe: 19. Discrepancies were resolved by a third independent reviewer. Inter-rater agreement was assessed using Cohen’s Kappa coefficient [15], which yielded a value of 0.56 (agreement rate: 84.36%), interpreted as moderate agreement.
2.2. Inclusion and Exclusion Criteria
Inclusion criteria: original, peer-reviewed, full-text articles published in scientific journals within the last 10 years, specifically in the period from 2015 to 2025 with an available abstract, focusing on the assessment of cerebrospinal fluid (CSF) volume and distribution in head CT images using artificial intelligence (AI).
Exclusion criteria: systematic reviews, meta-analyses, conference abstracts, congress proceedings, posters, preprints, letters, studies involving paediatric populations, and any publication types other than original research articles. Studies utilizing imaging modalities other than CT (e.g., MRI); studies older than 10 years; studies not involving AI; studies not focused on head CT; and studies in which CSF was not measured using AI techniques.
2.3. Data Extraction
All articles meeting the inclusion criteria underwent data extraction by three independent reviewers. From each study, the following information was collected:
General characteristics: authors, country of origin, publication year, journal name, impact factor (IF), overall aim of the study, and keywords.
Dataset information: number of unique patients, mean patient age, and number of CT scans assigned to training, validation, and testing sets.
Artificial Intelligence (AI): characteristics of the AI method used, software utilized for segmentation, and details of image preprocessing.
Study type: prospective or retrospective.
Results: Dice Similarity Coefficient (DSC) and correlation coefficients (r) of volumetric measurements.
2.4. Quality Assessment
This systematic review was conducted in accordance with the PRISMA 2020 Statement (Supplement S3) and PRISMA 2020 for Abstracts (Supplement S4) [16]. The quality of the included studies was assessed using the following tools: the JBI Checklist for systematic reviews (Supplement S5) [17], AMSTAR 2 (Supplement S6)—scale for critically appraising systematic reviews of healthcare interventions [18] and the Critical Appraisal Skills Programme (CASP)—diagnostic study checklist (Supplement S7) [19].
2.5. Risk of Bias Assessment
We applied the QUADAS-2 framework [20] to systematically evaluate methodological quality and potential bias in each included study. Four core domains were assessed—Patient Selection, Index Test, Reference Standard, and Flow and Timing—alongside three applicability dimensions to ensure alignment with our focus on AI-driven CSF segmentation in head CT. Two reviewers independently completed a tailored QUADAS-2 form, which included signaling questions specific to algorithmic segmentation methods. Disagreements were discussed until consensus was reached, and a third reviewer adjudicated any remaining conflicts. Final judgments (Low, High, or Unclear) for each domain are presented in [Supplement S8], providing a transparent basis for interpreting our review’s conclusions.
3. Results
3.1. Purpose of Research
All articles included in this systematic review used AI for CSF segmentation and analysis. However, they focused on different aspects and had different objectives. The detailed objective of each study is presented in Table 1.
3.2. Sample Size and Mean Age of Study Populations
The number of unique patients varied considerably across the included studies. The smallest cohort, described by Andrei Irimia et al., comprised 35 participants [26], while the largest, reported by Rajat Dhar et al., included 738 subjects [31]. The mean age of patients also differed between studies. The lowest average age was reported by James Booker et al. (55 ± 7 years) [29], whereas the highest was observed in the study by Liang Chen et al. (71 ± 14 years) [30]. One study did not report any information regarding the mean age of its patient population [33].
3.3. Type of Study
Nearly all of the studies included in this systematic review were retrospective in nature [21,22,23,24,25,26,27,28,30,31,32,33,34]. Only one study, conducted by James Booker et al., followed a prospective design [29].
3.4. Characteristics of the AI Technology
The reviewed studies employed various artificial intelligence models, primarily based on machine learning. The most frequently used approaches included Random Forest algorithms [21,25,28,29] and Convolutional Neural Networks (CNN) [22,23,24,27,31]. The most commonly utilized software for image segmentation was SPM12 [23,24,26] and MIPAV [25,29,31]. Three studies did not report the specific software used for segmentation [21,32,34]. A detailed summary of the types of AI algorithms and segmentation software employed across the included studies is provided in Table 2.
3.5. Provided Results
Some authors did not provide detailed results about the accuracy of CSF segmentation by the trained AI models [24,28,31,34]. In contrast, Songsaeng et al. related their findings to the effectiveness of the AI model in diagnosing normal-pressure hydrocephalus [24]. The study by Dhar et al. reported results demonstrating high reproducibility of CSF volume measurements by AI across multiple scans (R = 0.98). Additionally, the AI model showed a significant correlation between increasing CSF volume and patient age (R = 0.73) [28]. In another study by Dhar et al., changes in CSF volume were analyzed as a biomarker for early brain edema development. The model’s performance in measuring CSF volume was also assessed by its correlation with patient age (R = 0.69) [31]. Foroushani et al. evaluated their model primarily based on its sensitivity and accuracy in diagnosing malignant brain edema. The models achieved high sensitivity values (up to 100%) but exhibited low accuracy rates (15–32%) [34].
Most authors assessed the AI’s effectiveness in automatic CSF segmentation by comparing the overlap between AI-generated segmentations and the manual CSF measurements performed by experts, typically quantified using the Dice Similarity Coefficient (DSC) [21,22,25,26,27,29,30,32,33]. The highest DSC values were reported in the studies by Dhar [21], Irimia et al. [26], and Liang et al. [30].
Three studies reported the correlation of volumetric measures between AI-based CSF volume estimates and manual measurements [23,25,32]. All these studies demonstrated high correlation coefficients, with the best result (r = 0.99) observed in the study by Yuan et al. [32]. This indicates a strong agreement between CSF volume measurements obtained via AI models and those derived from manual expert segmentation.
Two studies assessed the consistency and reproducibility of CSF volume measurements using the Intraclass Correlation Coefficient (ICC) [22,26]. Puzio et al. compared the agreement between AI-based automatic CSF volume measurements and manual CSF volume measurements [22]. Irimia et al. compared AI-derived automatic CSF volume measurements from head CT scans with those obtained from MRI scans using the FreeSurfer 6.0 software [23].
Across the included studies, AI-based CSF segmentation showed high performance, with DSC ranging from approximately 0.72 to 0.95, volumetric correlations (r) between 0.91 and 0.99, and ICC between 0.61 and 0.96. A summary of these results is provided in Table 3.
Quality assessment using JBI, AMSTAR 2, and CASP indicated that most studies were methodologically sound but limited by retrospective design, single-center cohorts, and incomplete reporting of validation. According to QUADAS-2, the overall risk of bias was judged as moderate, with the highest concerns in patient selection and reference standards. No quantitative meta-analysis was performed because of substantial heterogeneity in study designs, clinical populations, AI architectures, reference standards, and reported outcomes; results were therefore narratively synthesized.
4. Discussion
In this article, we present a systematic review concerning the use of AI for segmentation and volumetric assessment of cerebrospinal fluid (CSF) in CT imaging. Various AI models were used in the analyzed studies, predominantly Random Forest algorithms [21,25,28,29] and CNNs [22,23,24,27,31]. The reviewed studies focused on different clinical indications for CSF volume measurement, with ischemic stroke, brain edema, hydrocephalus, and general assessments of white/gray matter and CSF volumes being the most common.
The results presented in the studies (Table 3) include Dice Similarity Coefficient (DSC), Correlation of Volumetric Measures (Pearson’s correlation coefficient, r), and Intraclass Correlation Coefficient (ICC) [22,26]. DSC measures the degree of spatial overlap between the AI-generated segmentation and a reference segmentation (indicating segmentation accuracy), the correlation coefficient assesses volumetric agreement between methods, and ICC evaluates measurement consistency based on variance analysis. The highest DSC values were reported by Dhar, Liang et al., and Sil C et al., indicating highly precise segmentation [21,27,30]. Lower DSC values observed in studies by Puzio et al., Booker et al., and Chen et al. may reflect greater segmentation challenges, more complex cases, or less effective AI algorithms [22,25,29]. High volumetric correlation confirms strong agreement between AI and reference methods, even when DSC is moderate, as seen in the work of Srikrishna et al. [23]. Yuan et al. reported nearly perfect volumetric agreement (r = 0.99), suggesting excellent model calibration for volume estimation [32].
We used the ICC classification established by the researchers, which in the study conducted by Puzio et al. [22] showed high agreement (>0.9) [35]. Regarding DSC, all reviewed studies achieved values above 0.75, with the majority exceeding 0.8, commonly accepted as an acceptable threshold [36]. These results indicate that AI-based segmentation correlates strongly with the manual gold standard used in individual studies. Given the above 80% overlap of correctly segmented areas, AI technologies have the potential to assist clinicians in CSF segmentation on head CT scans in the future. Such automation could reduce the time required for image analysis and limit repetitive tasks currently performed by medical personnel.
Comparable systematic reviews have explored AI’s role in Adaptive Radiotherapy (ART) for head and neck cancers [37], AI-driven segmentation in Cone-Beam CT (CBCT) imaging of the jaw and mandible [38] and AI-assisted detection of intracerebral hemorrhages in non-contrast CT scans of patients with acute stroke [39]. These studies report DSC values ranging from 0.73 to 0.92, similar to those observed in our review. This suggests that AI applications are advancing not only in CSF analysis and segmentation but also across other areas of medical imaging diagnostics.
Therefore, it can be concluded that AI-based methods could in the future serve as supportive tools for clinicians in identifying pathological conditions by assessing CSF volume changes over time. However, clinical implementation should be deferred until these technologies undergo more rigorous testing and refinement. Although some studies have reported DSC values exceeding 0.9 for segmentation accuracy relative to manual standards, these results were generally obtained on relatively small patient cohorts (e.g., 38 [21], 35 [26], 133 [30] patients). A slightly larger group of patients was included in the study with the highest R-value (0.99) (244 patients) [32]. Additionally, as this review included only peer-reviewed full texts, a potential publication bias may be present.
To achieve reliable outcomes and enable clinical application, prospective, standardized studies in larger patient populations are required. We further recommend standardizing performance metrics across groups, leveraging Big Data to build larger and more diverse datasets, and clearly defining the clinical utility for specific neurological conditions. Finally, clinical adoption will depend on practical integration into routine workflows (PACS/RIS interoperability, predictable turnaround times) and compliance with regulatory pathways for software as a medical device.
5. Conclusions
This systematic review of 14 studies highlights CNNs and Random Forest models for automated segmentation and volumetric analysis of CSF on head CT scans. Across the included studies, AI-driven methods consistently achieved high segmentation accuracy (as demonstrated by high DSC) and strong agreement with manual measurements, indicating their potential in neuroradiology.
The current evidence is limited by relevant heterogeneity in datasets and algorithms, retrospective study designs, and relatively small patient cohorts, as well as a lack of prospective clinical validation. To translate these promising findings into routine practice, future research must focus on methodological standardization (including unified segmentation protocols and performance metrics), larger multi-center datasets, and well-designed prospective trials to confirm its utility. Finally, the integration of validated AI tools into everyday radiological workflows could accelerate detection of subtle CSF changes, support clinical decision-making, and enhance overall patient care in neuroimaging.
Conceptualization, M.B.; methodology, M.B. and A.M.; validation, M.B.; formal analysis, M.R.; investigation, M.B., A.M. and D.K.; data curation, M.R. and M.M.; writing—original draft preparation, M.B., A.M., D.K., M.M.; writing—review and editing, M.B., A.M., D.K., J.K. and S.G.; visualization, A.M.; supervision, J.K. and S.G.; project administration, M.B.; funding acquisition, M.B. All authors have read and agreed to the published version of the manuscript.
Not applicable.
Not applicable.
The original contributions presented in the study are included in the article/
The authors declare no conflicts of interest.
Footnotes
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Aims of the included studies.
| Authors | Aim of the Method |
|---|---|
| Rajat Dhar [ | Development of a neural network-based image segmentation algorithm that can automatically measure CSF volume on serial CT scans from stroke patients. |
| Tomasz Puzio et al. [ | Assessment on whether CSF distribution evaluated by a specifically developed deep-learning neural network (DLNN) could assist in quantifying mass effect. |
| Meera Srikrishna et al. [ | Development of an automatic method for segmenting grey matter (GM), white matter (WM), cerebrospinal fluid (CSF), and intracranial volume (ICV) from head CT images. |
| Dittapong Songsaeng et al. [ | Improvement of normal-pressure hydrocephalus diagnosis by comparing the accuracy, sensitivity, specificity, and predictive values of radiological parameters, as evaluated by radiologists and AI methods, utilizing cerebrospinal fluid volumetry. |
| Yasheng Chen et al. [ | Development and validation of an automated technique for CSF segmentation via integration of random forest (RF) based machine learning with geodesic active contour (GAC) segmentation. |
| Andrei Irimia et al. [ | Evaluation of AI performance in segmenting white matter, gray matter, and cerebrospinal fluid from head CT images. |
| Sil C. Van De Leemput et al. [ | Development of a fully convolutional neural network (CNN) for 3D multiclass segmentation in 4D head CT, trained end-to-end using sparse 2D annotations. |
| Rajat Dhar et al. [ | Development of a machine learning algorithm capable of segmenting and measuring CSF volume from serial CT scans of stroke patients. |
| James Booker et al. [ | Describing the relationship between blood and CSF volumes in different compartments on baseline CT after aSAH, assess whether they independently predict long-term outcomes with AI integration, and explore their interaction with age. |
| Liang Chen et al. [ | Development of a novel CNN architecture, called Dense-Res-Inception Net (DRINet), to improve feature extraction and enhance segmentation accuracy in medical images, particularly in cases where differences between categories are subtle in terms of intensity, location, shape, and size. |
| Rajat Dhar et al. [ | Assessment on whether early changes in cerebrospinal fluid volume, measured using a neural network-based algorithm, can serve as an early biomarker for cerebral edema and poor clinical outcomes in stroke patients. |
| Jane Y Yuan et al. [ | Development of an automated algorithm to extract selective sulcal volume (SSV) and evaluate the age-dependent relationship of reduced SSV on early outcomes after a SAH. |
| Kevin T. Huang et al. [ | Development of an algorithm that can automatically detect ventriculomegaly on head CT scans, serving as an indicator of shunt failure in real-life adult hydrocephalus patients. The algorithm aims to achieve this by accurately identifying the lateral and third ventricles. |
| Hossein Mohammadian Foroushani et al. [ | Development and evaluation of machine learning models, specifically fully connected and long short-term memory (LSTM) neural networks, to predict which stroke patients will require hemicraniectomy or die due to midline shift. These models use serial clinical and imaging data, including volumetric cerebrospinal fluid (CSF) measurements extracted from baseline and 24-h CT scans. |
Detailed characteristics of the AI and segmentation software.
| Authors | Characteristics of the AI Technology | Software Used for Segmentation |
|---|---|---|
| Rajat Dhar [ | Random forest model, which was refined by training a fully CNN (based on the U-Net architecture) to accurately perform the segmentation of CSF | - |
| Tomasz Puzio et al. [ | Convolutional neural network with basic U-Net architecture in a 3D version | Exhibeon3 DICOM viewer |
| Meera Srikrishna et al. [ | U-Net deep learning model | SPM12 |
| Dittapong Songsaeng et al. [ | Modified 2D U-Net model | SPM12 |
| Yasheng Chen et al. [ | Random forest classifiers using Haar-like features combined with geodesic active contour (GAC) refinement via the level-set method | MIPAV |
| Andrei Irimia et al. [ | Gaussian Mixture Model (GMM)-based segmentation approach adapted from probabilistic classification methods, incorporating topology-constrained segmentation inspired by Ashburner and Friston’s methods and refined using Bayesian inference principles and a priori tissue probability maps | SPM 12 and MATLAB |
| Sil C. Van De Leemput et al. [ | 3D CNN architecture, inspired by U-Net | VCAST (volumetric cluster annotation and segmentation tool) |
| Rajat Dhar et al. [ | Random forest machine learning | XNAT (eXtensible Neuroimaging Archive Toolkit) |
| James Booker et al. [ | Random forest machine learning | MIPAV |
| Liang Chen et al. [ | Dense-Res-Inception Net (DRINet) | MRICron |
| Rajat Dhar et al. [ | Convolutional neural network (based on the U-Net architecture) | MIPAV |
| Jane Y Yuan et al. [ | Deep learning-based approach: a four-layer U-Net | - |
| Kevin T. Huang et al. [ | Two-dimensional U-Net | 3D Slicer |
| Hossein Mohammadian Foroushani et al. [ | A fully connected neural network and a recurrent neural network that employed a Long Short-Term Memory (LSTM) architecture | - |
Results.
| Authors | Dice Similarity Coefficient (DSC) | Correlation of | Intraclass Correlation Coefficient (ICC) |
|---|---|---|---|
| Dhar [ | 0.95 | - | - |
| Puzio et al. [ | 0.782 (training); | - | 0.96 |
| Srikrishna et al. [ | 0.75 | 0.91 | - |
| Chen et al. [ | 0.751 ± 0.059 (baseline scans); | 0.92 | - |
| Irimia et al. [ | 0.92 ± 0.007 (study group)/ | - | 0.74 (study group)/ |
| Sil C. Van De Leemput et al. [ | 0.86 ± 0.04 | - | - |
| Booker et al. [ | 0.7604 ± 0.106 | - | - |
| Liang et al. [ | 0.92 | - | - |
| Yuan et al. [ | 0.82 ± 0.11 | 0.99 | - |
| Huang et al. [ | 0.809 ± 0.094 | - | - |
Supplementary Materials
The following supporting information can be downloaded at:
1. Wilson, M.H. Monro-Kellie 2.0: The Dynamic Vascular and Venous Pathophysiological Components of Intracranial Pressure. J. Cereb. Blood Flow. Metab.; 2016; 36, pp. 1338-1350. [DOI: https://dx.doi.org/10.1177/0271678X16648711] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/27174995]
2. Kim, D.-J.; Czosnyka, Z.; Kasprowicz, M.; Smieleweski, P.; Baledent, O.; Guerguerian, A.-M.; Pickard, J.D.; Czosnyka, M. Continuous Monitoring of the Monro-Kellie Doctrine: Is It Possible?. J. Neurotrauma; 2012; 29, pp. 1354-1363. [DOI: https://dx.doi.org/10.1089/neu.2011.2018] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/21895518]
3. Mitręga, A.; Kaczyńska, D.; Bielówka, M.; Rojek, M.; Janik, M.; Dudek, P.; Denisiewicz, N.; Wocław, A.; Czogalik, Ł.; Stencel, M.
4. Lipková, J.; Menze, B.; Wiestler, B.; Koumoutsakos, P.; Lowengrub, J.S. Modelling Glioma Progression, Mass Effect and Intracranial Pressure in Patient Anatomy. J. R. Soc. Interface; 2022; 19, 20210922. [DOI: https://dx.doi.org/10.1098/rsif.2021.0922]
5. Kufel, J.; Bargieł-Łączek, K.; Kocot, S.; Koźlik, M.; Bartnikowska, W.; Janik, M.; Czogalik, Ł.; Dudek, P.; Magiera, M.; Lis, A.
6. Gore, J.C. Artificial Intelligence in Medical Imaging. Magn. Reson. Imaging; 2020; 68, pp. A1-A4. [DOI: https://dx.doi.org/10.1016/j.mri.2019.12.006]
7. Uparela-Reyes, M.J.; Villegas-Trujillo, L.M.; Cespedes, J.; Velásquez-Vera, M.; Rubiano, A.M. Usefulness of Artificial Intelligence in Traumatic Brain Injury: A Bibliometric Analysis and Mini-Review. World Neurosurg.; 2024; 188, pp. 83-92. [DOI: https://dx.doi.org/10.1016/j.wneu.2024.05.065]
8. Yousef, R.; Khan, S.; Gupta, G.; Siddiqui, T.; Albahlal, B.M.; Alajlan, S.A.; Haq, M.A. U-Net-Based Models towards Optimal MR Brain Image Segmentation. Diagnostics; 2023; 13, 1624. [DOI: https://dx.doi.org/10.3390/diagnostics13091624]
9. Beeche, C.; Singh, J.P.; Leader, J.K.; Gezer, S.; Oruwari, A.P.; Dansingani, K.K.; Chhablani, J.; Pu, J. Super U-Net: A Modularized Generalizable Architecture. Pattern Recognit.; 2022; 128, 108669. [DOI: https://dx.doi.org/10.1016/j.patcog.2022.108669]
10. Li, Z.; Feng, N.; Pu, H.; Dong, Q.; Liu, Y.; Liu, Y.; Xu, X. PIxel-Level Segmentation of Bladder Tumors on MR Images Using a Random Forest Classifier. Technol. Cancer Res. Treat.; 2022; 21, 15330338221086395. [DOI: https://dx.doi.org/10.1177/15330338221086395]
11. Qamar, S.; Öberg, R.; Malyshev, D.; Andersson, M. A Hybrid CNN-Random Forest Algorithm for Bacterial Spore Segmentation and Classification in TEM Images. Sci. Rep.; 2023; 13, 18758. [DOI: https://dx.doi.org/10.1038/s41598-023-44212-5]
12. van Hal, S.T.; van der Jagt, M.; van Genderen, M.E.; Gommers, D.; Veenland, J.F. Using Artificial Intelligence to Predict Intracranial Hypertension in Patients After Traumatic Brain Injury: A Systematic Review. Neurocrit. Care; 2024; 41, pp. 285-296. [DOI: https://dx.doi.org/10.1007/s12028-023-01910-2]
13. Schardt, C.; Adams, M.B.; Owens, T.; Keitz, S.; Fontelo, P. Utilization of the PICO Framework to Improve Searching PubMed for Clinical Questions. BMC Med. Inform. Decis. Mak.; 2007; 7, 16. [DOI: https://dx.doi.org/10.1186/1472-6947-7-16] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/17573961]
14. Ouzzani, M.; Hammady, H.; Fedorowicz, Z.; Elmagarmid, A. Rayyan—A Web and Mobile App for Systematic Reviews. Syst. Rev.; 2016; 5, 210. [DOI: https://dx.doi.org/10.1186/s13643-016-0384-4] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/27919275]
15. McHugh, M.L. Interrater Reliability: The Kappa Statistic. Biochem. Med.; 2012; 22, pp. 276-282. [DOI: https://dx.doi.org/10.11613/BM.2012.031]
16. Page, M.J.; Moher, D.; Bossuyt, P.M.; Boutron, I.; Hoffmann, T.C.; Mulrow, C.D.; Shamseer, L.; Tetzlaff, J.M.; Akl, E.A.; Brennan, S.E.
17. Aromataris, E.; Fernandez, R.; Godfrey, C.M.; Holly, C.; Khalil, H.; Tungpunkom, P. Summarizing Systematic Reviews: Methodological Development, Conduct and Reporting of an Umbrella Review Approach. Int. J. Evid. Based Healthc.; 2015; 13, pp. 132-140. [DOI: https://dx.doi.org/10.1097/XEB.0000000000000055]
18. Shea, B.J.; Reeves, B.C.; Wells, G.; Thuku, M.; Hamel, C.; Moran, J.; Moher, D.; Tugwell, P.; Welch, V.; Kristjansson, E.
19. Long, H.A.; French, D.P.; Brooks, J.M. Optimising the Value of the Critical Appraisal Skills Programme (CASP) Tool for Quality Appraisal in Qualitative Evidence Synthesis. Res. Methods Med. Health Sci.; 2020; 1, pp. 31-42. [DOI: https://dx.doi.org/10.1177/2632084320947559]
20. Whiting, P.F.; Rutjes, A.W.S.; Westwood, M.E.; Mallett, S.; Deeks, J.J.; Reitsma, J.B.; Leeflang, M.M.G.; Sterne, J.A.C.; Bossuyt, P.M.M. QUADAS-2 Group. QUADAS-2: A Revised Tool for the Quality Assessment of Diagnostic Accuracy Studies. Ann. Intern. Med.; 2011; 155, pp. 529-536. [DOI: https://dx.doi.org/10.7326/0003-4819-155-8-201110180-00009]
21. Dhar, R. Automated Quantitative Assessment of Cerebral Edema after Ischemic Stroke Using CSF Volumetrics. Neurosci. Lett.; 2020; 724, 134879. [DOI: https://dx.doi.org/10.1016/j.neulet.2020.134879]
22. Puzio, T.; Matera, K.; Wiśniewski, K.; Grobelna, M.; Wanibuchi, S.; Jaskólski, D.J.; Bobeff, E.J. Automated Volumetric Evaluation of Intracranial Compartments and Cerebrospinal Fluid Distribution on Emergency Trauma Head CT Scans to Quantify Mass Effect. Front. Neurosci.; 2024; 18, 1341734. [DOI: https://dx.doi.org/10.3389/fnins.2024.1341734]
23. Srikrishna, M.; Pereira, J.B.; Heckemann, R.A.; Volpe, G.; van Westen, D.; Zettergren, A.; Kern, S.; Wahlund, L.-O.; Westman, E.; Skoog, I.
24. Songsaeng, D.; Nava-apisak, P.; Wongsripuemtet, J.; Kingchan, S.; Angkoondittaphong, P.; Phawaphutanon, P.; Supratak, A. The Diagnostic Accuracy of Artificial Intelligence in Radiological Markers of Normal-Pressure Hydrocephalus (NPH) on Non-Contrast CT Scans of the Brain. Diagnostics; 2023; 13, 2840. [DOI: https://dx.doi.org/10.3390/diagnostics13172840] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37685378]
25. Chen, Y.; Dhar, R.; Heitsch, L.; Ford, A.; Fernandez-Cadenas, I.; Carrera, C.; Montaner, J.; Lin, W.; Shen, D.; An, H.
26. Irimia, A.; Maher, A.S.; Rostowsky, K.A.; Chowdhury, N.F.; Hwang, D.H.; Law, E.M. Brain Segmentation From Computed Tomography of Healthy Aging and Geriatric Concussion at Variable Spatial Resolutions. Front. Neuroinf.; 2019; 13, 9. [DOI: https://dx.doi.org/10.3389/fninf.2019.00009] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30936828]
27. Van De Leemput, S.C.; Meijs, M.; Patel, A.; Meijer, F.J.A.; Van Ginneken, B.; Manniesing, R. Multiclass Brain Tissue Segmentation in 4D CT Using Convolutional Neural Networks. IEEE Access; 2019; 7, pp. 51557-51569. [DOI: https://dx.doi.org/10.1109/ACCESS.2019.2910348]
28. Dhar, R.; Chen, Y.; An, H.; Lee, J.-M. Application of Machine Learning to Automated Analysis of Cerebral Edema in Large Cohorts of Ischemic Stroke Patients. Front. Neurol.; 2018; 9, 687. [DOI: https://dx.doi.org/10.3389/fneur.2018.00687]
29. Booker, J.; Zolnourian, A.; Street, J.; Arora, M.; Pandit, A.S.; Toma, A.; Wu, C.-H.; Galea, I.; Bulters, D. Quantification of Blood and CSF Volume to Predict Outcome after Aneurysmal Subarachnoid Hemorrhage. Neurosurg. Rev.; 2024; 47, 752. [DOI: https://dx.doi.org/10.1007/s10143-024-03001-y]
30. Chen, L.; Bentley, P.; Mori, K.; Misawa, K.; Fujiwara, M.; Rueckert, D. DRINet for Medical Image Segmentation. IEEE Trans. Med. Imaging; 2018; 37, pp. 2453-2462. [DOI: https://dx.doi.org/10.1109/TMI.2018.2835303]
31. Dhar, R.; Chen, Y.; Hamzehloo, A.; Kumar, A.; Heitsch, L.; He, J.; Chen, L.; Slowik, A.; Strbian, D.; Lee, J.-M. Reduction in CSF Volume as an Early Quantitative Biomarker of Cerebral Edema after Ischemic Stroke. Stroke; 2020; 51, pp. 462-467. [DOI: https://dx.doi.org/10.1161/STROKEAHA.119.027895] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31818229]
32. Yuan, J.Y.; Chen, Y.; Kumar, A.; Zlepper, Z.; Jayaraman, K.; Aung, W.Y.; Clarke, J.V.; Allen, M.; Athiraman, U.; Osbun, J.
33. Huang, K.T.; McNulty, J.; Hussein, H.; Klinger, N.; Chua, M.M.J.; Ng, P.R.; Chalif, J.; Mehta, N.H.; Arnaout, O. Automated Ventricular Segmentation and Shunt Failure Detection Using Convolutional Neural Networks. Sci. Rep.; 2024; 14, 22166. [DOI: https://dx.doi.org/10.1038/s41598-024-73167-4] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/39333724]
34. Foroushani, H.M.; Hamzehloo, A.; Kumar, A.; Chen, Y.; Heitsch, L.; Slowik, A.; Strbian, D.; Lee, J.-M.; Marcus, D.S.; Dhar, R. Accelerating Prediction of Malignant Cerebral Edema after Ischemic Stroke with Automated Image Analysis and Explainable Neural Networks. Neurocrit. Care; 2022; 36, pp. 471-482. [DOI: https://dx.doi.org/10.1007/s12028-021-01325-x]
35. Koo, T.K.; Li, M.Y. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J. Chiropr. Med.; 2016; 15, pp. 155-163. [DOI: https://dx.doi.org/10.1016/j.jcm.2016.02.012]
36. Brock, K.K.; Mutic, S.; McNutt, T.R.; Li, H.; Kessler, M.L. Use of Image Registration and Fusion Algorithms and Techniques in Radiotherapy: Report of the AAPM Radiation Therapy Committee Task Group No. 132. Med. Phys.; 2017; 44, pp. e43-e76. [DOI: https://dx.doi.org/10.1002/mp.12256]
37. Mastella, E.; Calderoni, F.; Manco, L.; Ferioli, M.; Medoro, S.; Turra, A.; Giganti, M.; Stefanelli, A. A Systematic Review of the Role of Artificial Intelligence in Automating Computed Tomography-Based Adaptive Radiotherapy for Head and Neck Cancer. Phys. Imaging Radiat. Oncol.; 2025; 33, 100731. [DOI: https://dx.doi.org/10.1016/j.phro.2025.100731]
38. Badr, F.F.; Jadu, F.M. Performance of Artificial Intelligence Using Oral and Maxillofacial CBCT Images: A Systematic Review and Meta-Analysis. Niger. J. Clin. Pract.; 2022; 25, pp. 1918-1927. [DOI: https://dx.doi.org/10.4103/njcp.njcp_394_22]
39. Hu, P.; Yan, T.; Xiao, B.; Shu, H.; Sheng, Y.; Wu, Y.; Shu, L.; Lv, S.; Ye, M.; Gong, Y.
© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.