RPS 1111 - Neuro-oncology and spine disorders: cutting-edge MRI approaches

Author Block: L. Joo, J. B. Lee, Y. Yim, N. H. Kim; Seoul/KR
Purpose: The purpose of this study is to characterize ISE observed in patients undergoing brain MRI for the evaluation of brain metastasis
Methods or Background: This retrospective study included patients who underwent enhanced brain MRI between November, 2023 and September, 2024 for the purpose of evaluating brain metastasis. All patients underwent CE-T2FLAIR, CE-3D MPRAGE, and CE-3D SPACE imaging. The obtained contrast-enhanced T2FLAIR images were evaluated by two independent radiologists to identify patients with sulcal enhancement, and in case of disagreement, a consensus conclusion was reached. The patients were then divided into true leptomeningeal metastasis and ISE based on imaging and clinical findings, and the incidence of ISE was investigated, and logistic regression analysis was performed to identify variables with statistically significant associations with risk factors such as age, microbleed (3 categories; 0, 1-4, 5≤), and previous brain radiation therapy (RT) history.
Results or Findings: A total of 387 patients who suspected malignancy underwent brain MRI, and 50 were excluded, resulting in 337 brain MRIs included in this analysis. Among these, 72 patients showed sulcal enhancement on CE-T2FLAIR. Of these, 11 were diagnosed as true leptomeningeal metastasis, and 61 were classified as ISE, yielding an incidence of 18.1% (61/337). Logistic regression identified age, microbleed, and previous RT history as significant risk factors. Multivariate logistic regression results showed that age, microbleed, and RT history were independent risk factors to predict ISE on CE-T2FLAIR, regardless of gender. (Odds ratio; age: 1.1, 1.0 ~ 1.1; microbleeds: 1-4: 2.9, 1.5 ~ 5.8; ≥5: 7.3, 2.6 ~ 21.0; RT: 7.7, 2.1 ~ 28.0)
Conclusion: The incidence of ILE performed to evaluate brain metastasis was 18.1% . Significant independent risk factors for detecting ILE include age, microbleed, and RT history.
Limitations: Not applicable
Funding for this study: The study was supported by grant from the Central Medical Service (CMS) Co., Ltd. Research Fund.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Chungang University Hospital 2501-016-19560

Author Block: C. García¹, I. MARTINEZ-ZALACAIN², P. Naval-Baudin², D. Choque², A. M. Viveros Castaño², M. Cos Domingo², C. Majós², A. Pons Escoda²; ¹Murcia/ES, ²Hospitalet de Llobregat/ES
Purpose: The aim of this study is to quantify the diagnostic performance of a multiparametric, bicompartmental perfusion analysis, and to establish technical-scenario-specific thresholds and classifiers for the presurgical differentiation between glioblastomas and metastases.
Methods or Background: Patients: Retrospective cohort with 101 glioblastomas IDH-wt (52 males, median age 61) and 52 metastases (29 males, median age 63). DSC-Perfusion: Double-acquisition protocol: 1st without pre-bolus, flip angle 75°; 2nd with pre-bolus (using prior contrast), flip angle 60°. Segmentations: Semi-automatic and 3D. Two tumor compartments: enhancing tumor and peritumoral T2-FLAIR abnormality (edema). Masks co-registered to DSC space. Perfusion metrics: Commercial AIF-based Bayesian vascular model. Voxel-wise and normalized to normal-appearing white matter. Macrovascular metrics: rCBV (relative-cerebral-blood-volume), rCBF (relative-cerebral-blood-flow), MTT (mean-transit-time). Microvascular: CTH (capillary-transit-time-heterogeneity), TTD (delay), COV (coefficient-of-variance). Metabolic: OEF (oxygen-extraction-fraction), rCMRO2 (relative-cerebral-metabolic-rate-of-oxygen). Data analysis: Computation of best single-metric/single-statistic performance per protocol and tumor compartment. Combination of metrics using logistic regressions on three-metric sets that maximized discrimination for each acquisition scenario and compartment, as well as bicompartmental models.
Results or Findings: Single-metric performance: The highest AUC was 0.78 for peritumoral rCBFmax in the second protocol. Remaining best single-metric AUCs: protocol-1/enhancing CTHp25 = 0.73; protocol-1/edema rCBFmax = 0.77; protocol-2/enhancing MTTmin = 0.74. Trivariable classifiers: Scenario-specific peak AUCs were protocol-2/enhancing 0.79 (MTTmin + TTDmean + rCMRO2mean); protocol-2/edema 0.79 (CTHmin + OEFmin + rCBFmax). Within protocols, bicompartmental models outperformed single-compartment models—protocol-1 0.82 (vs 0.75); protocol-2 0.80 (vs 0.79). All models ranged 0.75–0.83. In exploratory subanalysis, both single metrics and classifiers remained stable in lab-simulated heterogeneous settings, with AUCs 0.73–0.84.
Conclusion: Multiparametric bicompartmental perfusion analysis using innovative commercial software demonstrates excellent discriminatory power and robustness across technical settings. Strengthening biological plausibility, classifiers combining vascular and metabolic information from distinct tumor habitats achieved the highest performance.
Limitations: Single-centre, retrospective design.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study was approved by the Research Ethics Committee of Hospital Universitari de Bellvitge.

Author Block: Y. Jia, M. Yuan; Nanjing/CN
Purpose: Leptomeningeal metastasis (LM) is a devastating complication of advanced lung adenocarcinoma (LUAD), with high prevalence in EGFR-mutant patients. Imaging and prognostic differences among EGFR subtypes remain poorly understood, especially in those receiving intraventricular chemotherapy via Ommaya reservoir. This study investigated MRI features and survival outcomes across EGFR mutation subtypes in this context.
Methods or Background: In this multicenter retrospective study, LUAD patients with cerebrospinal fluid (CSF)-confirmed LM (November 2021–June 2024) underwent contrast-enhanced brain MRI and CSF-based next-generation sequencing (NGS). Patients were classified as EGFR exon 21 L858R (EGFR 21), exon 19 deletion (EGFR 19), or non-classic/other mutations (NCOM). Imaging features were compared using chi-square tests, and intracranial progression-free survival (iPFS) and overall survival (OS) were analyzed with Kaplan–Meier and Cox regression models.
Results or Findings: A total of 110 patients (mean age 55.6 ± 10.2 years) were included. MRI-negative LM was more frequent in EGFR 19 (65.5%) versus EGFR 21 (34.9%) and NCOM (34.2%) (p = 0.01). EGFR 21 was associated with localized disease (<4 lobes: 85.7%) compared with NCOM (≥4 lobes: 45.5%) (p = 0.01). Median iPFS and OS were longer in EGFR 21 (12.0 and 18.0 months) than in NCOM (6.5 and 10.2 months). EGFR 19 conferred no survival benefit. EGFR subtype independently predicted iPFS (p = 0.03), but not OS (p = 0.12).
Conclusion: EGFR mutation subtypes are linked to distinct MRI features and survival outcomes in LUAD patients with LM treated via Ommaya reservoir, providing insights for diagnosis, risk stratification, and treatment planning.
Limitations: This study is limited by modest sample size and lack of CSF biomarker correlation. The high proportion of MRI-negative LM in EGFR 19 underscores the need for multimodal imaging and molecular integration in future multicenter studies.
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: he First Affiliated Hospital of Nanjing Medical University Ethics Review Board

Author Block: Y. Liang, Z. Chen, M. Ge, Y. Wang; Beijing/CN
Purpose: Differentiating radiation-Induced brain Injury (RIBI) from tumor recurrence (TR) in brain metastases after stereotactic radiosurgery (SRS) remains a significant clinical challenge due to their similar appearance on conventional MRI. As management strategies differ substantially, this study aims to develop a deep learning model integrating multiparametric MRI with clinical data for accurate discrimination.
Methods or Background: This retrospective study enrolled 82 patients with brain metastases treated with SRS between January 2016 and December 2024. Diagnosis was confirmed by pathology or ≥6 months of follow-up. Based on post-radiation evolution, lesions were classified as RIBI (49 lesions) if showing initial growth followed by spontaneous regression or stability >6 months, or as TR (43 lesions) if demonstrating continuous progression. Multiparametric MRI sequences (T1WI, contrast-enhanced T1WI, T2WI) were analyzed, with all lesions manually delineated by radiologists. A deep learning model was developed to fuse imaging features with embedded clinical data (including patient demographics, tumor characteristics, and treatment parameters) for end-to-end classification. Model performance was evaluated via cross-validation using accuracy, sensitivity, specificity, and AUC.
Results or Findings: The integrated model demonstrated high performance, achieving an accuracy of 91.30%, sensitivity of 84.21%, specificity of 97.14%, and AUC of 0.92. Compared to the MRI-only model (accuracy 89.13%, AUC 0.885), clinical data integration improved accuracy by 2% and AUC by 0.035, validating the added value of multi-modal fusion.
Conclusion: The deep learning model integrating MRI and clinical data shows significant potential for accurately differentiating RIBI from TR in post-SRS brain metastases, providing a high-precision, non-invasive tool for clinical decision-making that may help avoid unnecessary invasive procedures.
Limitations: Not applicable
Funding for this study: Not applicable
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: V. Sawlani¹, N. Mei², R. Flintham¹, S. Meade¹, H. Benghiat¹, S. Nagaraju¹, U. Pohl¹, P. Sanghera¹, V. Wykes¹; ¹Birmingham/UK, ²Shanghai/CN
Purpose: Stereotactic radiosurgery (SRS) is widely used for brain metastases, but differentiating tumour progression from radiation necrosis on conventional MRI remains difficult. Percentage signal recovery (PSR), derived from dynamic susceptibility contrast (DSC) perfusion MRI, reflects signal recovery post-contrast and offers insights into capillary permeability. This study aimed to evaluate PSR and relative cerebral blood volume (rCBV) and assess their combined diagnostic value in post-SRS brain metastases.
Methods or Background: Patients with enlarging post-SRS brain metastases and diagnostic uncertainty were retrospectively included. PSR and rCBV were extracted from DSC-MRI and normalized to contralateral white matter. The dataset was split into training and validation cohorts using stratified sampling. Logistic regression with 5-fold cross-validation and bootstrap validation was used. Diagnostic performance was assessed by ROC analysis.
Results or Findings: Sixty-one patients (62 lesions; 26 progression, 36 necrosis) were included. Progression showed higher rCBV (2.84 vs. 0.76) and lower PSR (95% vs. 176%) (both p < 0.001). Both were significant in univariate analysis; PSR remained independently predictive (p = 0.04) in multivariate analysis. PSR outperformed rCBV in ROC analysis (AUC = 0.960 vs. 0.898); the combined model improved accuracy (95.9%) without loss of sensitivity or specificity. Bootstrap-derived thresholds were 108% (PSR) and 1.96 (rCBV). A nomogram was developed for individualized risk estimation.
Conclusion: PSR and rCBV provide complementary diagnostic information for post-SRS lesion assessment. PSR offers added value without additional scanning, and integration of both parameters enhances diagnostic confidence. Routine inclusion of PSR and rCBV in post-SRS imaging protocols is recommended.
Limitations: This retrospective single-centre study may be subject to selection bias and limited generalizability. Using maximum rCBV and PSR values likely improved sensitivity for detecting focal tumour recurrence but may introduce variability in heterogeneous lesions; prospective multicentre validation is warranted.
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Institutional Review Board of Queen Elizabeth Hospital Birmingham

Author Block: Z. H. Chen Zhou, A. Hilario Barrio, E. Salvador Alvarez, A. Cardenas, J. Romero Coronado, C. Lechuga Vázquez, A. Martínez De Aragón Calvo, R. D´Ambrosi, A. Ramos Gonzalez; Madrid/ES
Purpose: Stereotactic radiosurgery (SRS) is the treatment of choice for local control of isolated brain metastases. However, after SRS, it is common for metastases to show increased enhancement, making it difficult to distinguish between tumour progression and treatment-related effects. DSC perfusion is useful for evaluating haemodynamic changes, with rCBV being the most extensively studied parameter. Nevertheless, the percentage signal recovery (PSR), which is an indicator of blood–brain barrier integrity, has been assessed less frequently. This study aimed to compare the diagnostic performance of rCBV and PSR.
Methods or Background: A retrospective analysis was performed on patients with brain metastases who were treated with SRS between January 2020 and December 2024. The patients showed increased post-treatment enhancement and had available DSC perfusion. The final diagnosis of progression or radionecrosis was established by either surgical resection or clinical–radiological follow-up. Images were processed to obtain rCBV and PSR values. Statistical tests (Mann–Whitney U and Student's t-test) were applied to compare the two groups.
Results or Findings: A total of 79 treated brain metastases that subsequently grew were included in the analysis (45 cases of radionecrosis and 34 cases of progression). rCBV was significantly higher in cases of tumour progression (p < 0.001), while PSR was significantly lower (p < 0.001). Using a cut-off value of 1.7, rCBV achieved a sensitivity of 91.2%, a specificity of 60%, and an area under the curve (AUC) of 0.87. For PSR, the cutoff value was 70%. Sensitivity was 94.1%, specificity was 97.8%, and the AUC was 0.99.
Conclusion: Although traditionally overlooked, PSR is more accurate than rCBV at differentiating tumour progression from radionecrosis after radiosurgery. This establishes PSR as a highly useful diagnostic tool in clinical practice.
Limitations: Retrospectiva and single center study.
Funding for this study: No funding.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The hospital's local ethics committee affirms that it complies with ethical standards and good clinical practice.

Author Block: N. Mei¹, Y. Chen², V. Sawlani³, X. Li¹, J. Cui¹, Z. Zheng⁴, D. Wang¹, Y. Lu¹, B. Yin¹; ¹Shanghai/CN, ²Jiangsu/CN, ³Birmingham/UK, ⁴Shandong/CN
Purpose: Tumor growth imposes a considerable psychological impact on patients with incidental asymptomatic meningiomas. This study aimed to identify clinical, semantic, and multiparametric MRI features associated with growth potential and to develop a machine learning model for risk prediction, thereby informing personalized surveillance and management strategies.
Methods or Background: This retrospective multi-center study enrolled adult patients with incidental asymptomatic meningiomas confirmed by routine MRI. Tumors were manually segmented on CE-T1WI images. Radiomics features were extracted from CE-T1WI, T2-FLAIR, and ADC images and selected using correlation and Cox regression analyses. A random survival forest model was developed with five-fold cross-validation to predict tumor growth. Model performance was assessed by C-index and time-dependent ROC curves. Risk stratification was evaluated using Kaplan-Meier analysis.
Results or Findings: 421 patients with incidental asymptomatic meningiomas from Institution A were randomly split into training, validation, and testing sets, with an independent external validation set comprising 39 patients from Institutions B and C. Eleven significant predictors were incorporated into a random survival forest model, which demonstrated strong performance with C-indices of 0.928, 0.874, 0.872, and 0.860 in the training, validation, testing, and external validation sets, respectively. The model achieved consistently high time-dependent AUCs (> 0.80) at 1-, 2-, 3-, and 5-year follow-up, and stratified patients into significantly low and high growth-risk groups on Kaplan-Meier analysis.
Conclusion: Our MRI-based machine learning model reliably predicts growth risk in incidental asymptomatic meningiomas, enabling personalized surveillance. This may improve clinical decision-making and reduce unnecessary interventions and, importantly, patient anxiety.
Limitations: Patient inclusion was based on meningiomas identified at initial MRI, which may have led to rare misclassification of solitary fibrous tumor; however, follow-up likely minimized this risk. Only reproducible, interpretable radiomics features were included to reduce redundancy and overfitting.
Funding for this study: This work is sponsored by the Explorers Program of Shanghai (Grant no. 24TS1410800) and the National Natural Science Foundation of China (82281966).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Institutional Review Board of Huashan Hospital

Author Block: H-H. Chen, F-C. Chang; New Taipei City/TW
Purpose: Intracranial solitary fibrous tumors(SFT) are rare hypervascular dural based tumor, which showed overlapping imaging appearance with angiomatous meningioma(AM), another hypervascular dural based tumor. This study aim to differentiate the two different tumor on a routine MRI sequence.
Methods or Background: We retrospectively reviewed the pathology database at VGHTPE from January 2015 to September 2025, and included 20 cases of SFT and AM respectively. Preoperative MRI features was analyzed. Group comparisons were performed using chi-square or t-tests.
Results or Findings: SFT typically demonstrated T1 and T2 isointensity, while AM more often showed T1 hypointensity and T2 hyperintensity. Both tumors were hypervascular with marked enhancement and no significant restricted diffusion overall. Although AM more frequently exhibited mild restricted diffusion in peripheral region. The presence of peripheral restricted diffusion showed a sensitivity of 50%, specificity of 85%, and overall accuracy of 67.5% in identifying AM.
Conclusion: AM and SFT may exhibit overlapping imaging features but differ markedly in biological behavior and prognosis. Accurate differentiation is essential for guiding appropriate therapeutic strategies. Peripheral restricted diffusion appears to be a promising imaging marker for distinguishing AM from SFT.
Limitations: Limitations include small sample size and retrospective single-center design. Possible bias due to manual drawing and visual positioning of the ROI.
Funding for this study: Not applicable
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: T. Orhan, M. Oral, Z. Firat, G. Ekinci; Istanbul/TR
Purpose: To investigate the relationship between magnetic resonance (MR) elastography-derived stiffness values and diffusion (ADC) as well as conventional MR signal intensity ratios (SIR) in meningiomas
Methods or Background: We retrospectively analyzed 40 patients with histologically proven meningiomas who underwent preoperative MRI, including elastography. Elastography stiffness scores were compared with ADC values and SIRs from T1, T2, and FLAIR sequences.
Results or Findings: Elastography scores showed a significant negative correlation with ADC values (p<0,01), indicating that stiffer tumors tend to have lower ADC. In contrast, no significant associations were found between elastography and SIR values on T1, T2, or FLAIR sequences (p > 0.05 for all).
A secondary finding was noted: in the subgroup of patients with elevated T2 SIR (>1) but reduced FLAIR SIR (<1), elastography scores were higher (median = 4) compared to other patients (median = 3). This difference was statistically significant (p<0.05).
Conclusion: Preoperative elastography stiffness is inversely related to ADC in meningiomas, suggesting complementary roles of these imaging biomarkers. While conventional SIR values alone did not correlate with stiffness, the observation of high elastography scores in tumors with a “T2/FLAIR mismatch” may represent a promising secondary marker. Further studies with larger cohorts are warranted to validate this novel finding.
Limitations: Lack of direct correlation between the elastography stiffness values and a gold standard
Retrospective study
Small sample size
Single centred study
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The institutional review board approved this retrospective study (Yeditepe University, IRB #E.83321821-805.02.03-624, 18.04.2025).

Author Block: B. Akgun, E. Gülay, B. Baysal; Istanbul/TR
Purpose: Spondylolisthesis is defined as the anterior displacement of one vertebra relative to the caudal vertebra.This study aimed to evaluate the relationship between facet tropism and disc and facet joint degeneration in degenerative lumbar spondylolisthesis and to identify potential radiographic risk factors by comparison with healthy controls.
Methods or Background: Seventy patients with DLS at L3/4, L4/5, and L5/S1 and seventy controls were included.Lumbar MR images were independently assessed by two radiology resident.Intervertebral disc degeneration was graded using the Pfirrmann system, while facet joint degeneration was evaluated with Weishaupt and Fujiwara scores.Right and left facet angles and tropism were recorded.The reference plane was defined as a line through the disc midline and base of the spinous process;the facet line connected the anteromedial and posteromedial borders of the superior articular facet.Facet angle was calculated between the reference plane and facet line. Tropism was defined as an asymmetry ≥8°. Inter-observer agreement was assessed with Cohen’s Kappa; correlations were analyzed using Spearman’s test.
Results or Findings: Among patients,65.5% were male and 34.5% female. DLS involvement was 43.7% at L4-5, 40.8% at L5-S1, and 15.5% at L3-4.Pfirrmann scores at L5-S1 were significantly higher in DLS than controls(p<0.05), while L3-4 and L4-5 showed no difference.Weishaupt and Fujiwara scores at L4-5 were significantly higher in DLS(p<0.05); other levels showed no difference.Facet tropism prevalence did not differ significantly. Right-sided facet angle negatively correlated with Weishaupt and Fujiwara scores(p<0.05); no significant left-sided correlation was observed.At L3-4 and L5-S1, tropism was associated with higher degeneration scores(p<0.01).
Conclusion: Facet joint degeneration, along with facet morphology, may play important roles in DLS development.These findings suggest that facet morphology and tropism may be linked to degeneration and could serve as potential radiographic risk factors in DLS.
Limitations: No limitations were identified.
Funding for this study: No funding was provided for this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Istanbul Medeniyet University non-interventional health research ethics committee

Author Block: K-C. Lin, Y. Huang; Taipei City/TW
Purpose: Intraoperative MRI (ioMR) is an underutilized examination for facilitating resection of intramedullary spinal cord tumors. In the current case series, the authors present their technical and interpretation-related experiences with ioMR.
Methods or Background: A total of 15 patients who underwent ioMR for intramedullary spinal cord tumor resection from April to December 2024 were retrospectively included in this single-center series. Unique intraoperative imaging findings were evaluated (intraoperative contrast leakage [IOCL], blood product in the surgical cavity, residual high T2 signal in the spinal cord, and suspicious nodular enhancement), and ioMR imaging quality was assessed on a Likert scale. Technical details were recorded for ioMR sequences.
Results or Findings: The average scanning time was 58.9 ± 8.7 minutes. Residual tumors were identified in eight patients (53.3%), and two patients (13.3%) underwent additional resection. IOCL was the most prevalent imaging finding, observed in 73.3%–80.0% of patients, with good interobserver agreement. The median imaging quality was acceptable (Likert scale = 3), as assessed by two reviewers, with thoracic spine images exhibiting the worst quality among all segments.
Conclusion: Total resection can be achieved in patients with intramedullary spinal cord tumors under the assistance of ioMR. Radiologists should note unique imaging findings that have been previously identified, with particular attention given to IOCL, to correctly interpret intraoperative spine MRI.
Limitations: First, the relatively small number of patients with heterogeneous tumor pathology and surgical goals may render our experiences non-generalizable experience to all patients undergoing intramedullary tumor resection. Second, the average ioMR scan time was approximately 1 hour, and this time increased with patient transfer. Whether the benefits of ioMR for surgical outcomes justify the longer operating time and greater resources expended should be examined in further studies.
Funding for this study: Nil
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: L. Piorecki¹, M. Socha¹, J. Verdú-Díaz², J. Diaz-Manera², V. Straub², R. Tupler³, J. Polanska¹; ¹Poland/PL, ²Newcastle/UK, ³Modena/IT
Purpose: Neuromuscular disorders (NMDs) are genetically heterogeneous and often display subtle, visually indistinct MRI patterns. Automated stratification and classification may facilitate diagnosis, enable patient grouping, and enhance understanding of genotype–phenotype associations in NMD.
Methods or Background: The original dataset consisted of 1,351 T1-weighted MRI scans of the lower limbs from 616 patients with 9 different genetically confirmed neuromuscular disorders, collected within the MYO-GUIDE project between 2023 and 2025. The dataset was cleaned, and the MRI scans were preprocessed by removing skin, subcutaneous tissue and osseous structures. The thigh region in the MR volumes was found by referencing the femur. After cleaning, the data were partitioned into training, validation, and hold-out subsets, containing 139, 47, and 51 scans, respectively, with multiple timepoints of the same patient assigned exclusively to one subset. Three feature extraction approaches were investigated: two based on radiomic descriptors and one employing a convolutional autoencoder. Feature dimensionality was reduced using Boruta and CellBRF selection algorithms. Classifier optimization was performed using four-fold cross-validation with a grid search across Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), and Support Vector Machine (SVM) models, as well as their parameters, for each feature selection method. The best classifier was chosen based on the weighted AUC.
Results or Findings: For radiomic descriptors, Boruta outperformed CellBRF, whereas CellBRF performed better for autoencoder-derived features with higher dimensionality. The highest classification metrics were achieved for radiomics per pie slice feature space, reduced by Boruta and LR as the classifier. Achieving a weighted AUC metric of 0.87 and 0.90 for the validation and holdout subsets, respectively.
Conclusion: This approach suggests the potential of MRI-based automation to support differential diagnosis in neuromuscular disorders and contribute to the development of imaging biomarkers.
Limitations: Not applicable.
Funding for this study: This work was financed by CoMPaSS-NMD, Computational Models for new Patients Stratification Strategies of Neuromuscular Disorders, HORIZON RIA, Tools and Technologies for Healthy Society, ID: GAP-101080874
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: