RPS 1311 - Decoding the mind: sculpting neuroimaging with technology

Author Block: L. Leukert, A. Kronfeld, R. Paul, M. A. Brockmann, S. Altmann, A. Othman; Mainz/DE
Purpose: To evaluate the presence of motion artifacts in 1.5 T T1-weighted MRI scans using 3D motion correction via the Scout Accelerated Motion Estimation and Reduction (SAMER) framework versus conventional image reconstruction.
Methods or Background: MRI long scan times often cause motion artifacts, reducing image quality. SAMER uses an ultrafast pre-scan and repeated acquisition of a minimal number of additional k-space encoding lines, enabling feasible computation times. A preliminary study (14 volunteers) assessed SAMER’s effect on induced motion at 3T. The main study (82 patients) compared conventional resonstruction (Non-Moco) and SAMER (SAMER Moco) motion correction using 3D T1-weighted imaging at 1.5T. Radiologists evaluated images with a 5-point Likert scale.
Results or Findings: In the preliminary study, SAMER Moco showed significant improvements over Non-Moco across all imaging parameters (p < 0.001), with 52.4% and 66.7% of cases rated as excellent or good for artifact freedom and image quality, compared to 21.4% for Non-Moco. The main study underlined these findings. SAMER Moco demonstrated superior image quality and outperformed Non-Moco, particularly in diagnostic confidence and overall image quality (p < 0.0001). Diagnostic confidence was rated excellent or good in 93.8% of SAMER Moco cases versus 72.0% for Non-Moco. Similarly, 84.6% of SAMER Moco cases had excellent or good image quality (56.8% for Non-Moco). Odds ratios favoured SAMER Moco (5.444 and 5.807, respectively, p < 0.0001). Multi-reader agreement was excellent across all parameters.
Conclusion: The use of SAMER in T1-weighted imaging is feasible in clinical practice and significantly enhances the reliability of 1.5 T brain MRI by successfully mitigating motion artifacts.
Limitations: This study's limitations include its single-centre design, reliance on a single 3D MR sequence, and inclusion of both contrast-enhanced and non-contrast scans. SAMER may also be affected by patient-induced k-space gaps.
Funding for this study: This research received no funding from any public, commercial, or not-for-profit sources. The authors declare that they have no competing interests that are relevant for the content of this article.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This single-center prospective study was approved by our institution's local ethics committee, and written informed consent was obtained (approval number 2021-15811). Our study was conducted in accordance with the Declaration of Helsinki and its amendments.

Author Block: D. Fedders¹, A. Hellerbach², M. Eichner², C. Panknin³, S. Faby³, J. Wirths², V. Visser-Vandewalle², H. Treuer², S. Hunsche²; ¹Chemnitz/DE, ²Cologne/DE, ³Forchheim/DE
Purpose: Accurate determination of the rotational orientation of directional deep brain stimulation (DBS) electrodes is crucial for optimizing therapeutic outcomes in functional neurosurgery. Conventional CT methods, relying on artifact analysis, face precision limitations, especially at certain angles. Photon-counting detector CT (PCD-CT), with its superior resolution, offers a potential alternative. This study compares the efficacy of PCD-CT against conventional CT-based artifact analysis in determining DBS electrode orientation.
Methods or Background: A phantom study was conducted using directional leads from Boston Scientific, Medtronic, and Abbott embedded in cylindrical phantoms. The phantoms were scanned with PCD-CT for direct orientation detection and conventional CT for stripe artifact analysis. Scans covered varying polar angles to assess accuracy and consistency. Key metrics included orientation accuracy and dependency on lead position relative to the CT gantry.
Results or Findings: PCD-CT demonstrated high accuracy across all tested angles, independent of lead alignment. In contrast, conventional CT showed reduced precision, particularly at extreme angles where artifact detection was unreliable. PCD-CT enabled consistent, precise assessments of segmented contacts, enhancing postoperative DBS programming.
Conclusion: PCD-CT offers a robust solution for determining the rotational orientation of DBS electrodes, overcoming limitations of conventional artifact-based methods. This supports more accurate electrode positioning and programming, potentially improving functional neurosurgery outcomes.
Limitations: The phantom-based design may not replicate clinical complexity, limiting generalizability. The study only evaluated specific directional DBS leads, so results may not apply to other types. Additionally, PCD-CT’s limited availability could hinder immediate clinical application.
Funding for this study: None beside scanning time at the research facility from Siemens
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: Phantom study

Author Block: G. Lin, W. Chen, M. Chen, J. Ji; Lishui/CN
Purpose: This study aims to propose a multimodal magnetic resonance imaging (MRI)-based machine learning framework to effectively classify mild cognitive impairment (MCI) and no cognitive impairment (NCI) in patients with cerebral small vessel disease (CSVD).
Methods or Background: We enrolled 223 patients with CSVD, categorized into NCI (n = 121) and MCI (n = 102) groups based on neurocognitive assessments. Multimodal MRI data, including T1-weighted, resting-state functional MRI, and diffusion tensor images, were collected. Image preprocessing, feature extraction, and feature selection methods were applied to obtain MRI features from the three modalities. The AutoGluon platform was utilized for model development, and traditional machine learning algorithms were applied for comparison. The models were validated using a validation cohort of 97 patients with CSVD, and their performance was assessed via receiver operating characteristic curve (ROC) analysis.
Results or Findings: The AutoGluon model to distinguish MCI from NCI based on multimodal MRI features demonstrated a high area under the ROC curve (AUC), accuracy, sensitivity, specificity, and F1-score in the testing set (0.894, 85.65%, 84.31%, 86.78%, and 84.31%, respectively) and validation cohort (0.846, 79.38%, 81.82%, 77.36%, and 78.26%, respectively). Other models built using traditional machine learning algorithms had AUCs of 0.661–0.732, and their prediction accuracies were significantly lower than that of the AutoGluon model (P < 0.001).
Conclusion: Our study provides a multimodal MRI-based machine learning framework, utilizing the AutoGluon platform, that outperforms traditional algorithms in classifying MCI and NCI in patients with CSVD, offering a promising tool for the early prediction of MCI in CSVD.
Limitations: As a retrospective study, it is susceptible to selection bias, which may limit its generalizability.
Funding for this study: This study is supported by Zhejiang Public Welfare Research Program (LGF20H220002, LGF19H180010), and Zhejiang Provincial Healthcare Program (2024KY562)
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the Ethics Committee of the Fifth Affiliated Hospital of Wenzhou Medical University (approval number: 2024-266)

Author Block: C. Otgonbaatar¹, H. Kim², P-H. Jeon², S. H. Jeon², S. Cha², J-K. Ryu¹, H. Shim¹, S. M. Ko², J. Kim²; ¹Seoul/KR, ²Wonju-si/KR
Purpose: This study aimed to evaluate the diagnostic performance, image quality, and radiation dose among ultralow-dose protocol with deep learning reconstruction (DLR), ultralow-dose computed tomography (CT) with iterative reconstruction (IR), and conventional-dose protocols for detecting intracranial hemorrhage.
Methods or Background: This retrospective study enrolled 93 patients. All patients underwent follow-up noncontrast CT with ultralow-dose setting after initial conventional-dose CT within 5 days. A conventional-dose CT was obtained using 123–188 mA and IR. Ultralow-dose CT was obtained using 50 mA with IR and DLR. Qualitative assessments and quantitative assessments (image noise, differentiation between gray and white matter, and artifact) were conducted. The diagnostic performance for detecting intracranial hemorrhage using ultralow-dose CT with IR and ultralow-dose CT with DLR was assessed.
Results or Findings: An approximately 84.0% reduction in median volume CT dose index was found in the ultralow-dose CT protocol (5.6 mGy) compared with conventional-dose CT (35.02 mGy; IQR: 33.09–37.36). Ultralow-dose CT with DLR significantly (p < 0.001) improved image noise, SNR, and CNR compared with ultralow-dose CT with IR and conventional-dose CT. Ultralow-dose CT with DLR resulted in higher sensitivity (99.3% vs. 98.6%) and specificity (97.5% vs. 97.5%) for detecting intracranial hemorrhage than ultralow-dose CT with IR.
Conclusion: Ultralow-dose CT with DLR is an acceptable technique that provides higher image quality and diagnostic performance with a reduction in radiation dose of approximately 87.7% compared with conventional-dose CT.
Limitations: We did not investigate the effect of a tube current of <50 mA on the diagnostic performance of intracranial hemorrhage and image quality. Further validation is required to investigate a lower effective dose of <0.21 mSv. Additionally, all results were limited to one scanner, and acquisition parameters may require adjustment for different CT vendors.
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: No

Author Block: A. Zimny¹, W. N. Machaj¹, P. Podgórski¹, W. Fortuna¹, J. Huber², B. Bobek-Billewicz³, P. Tabakow¹; ¹Wrocław/PL, ²Poznań/PL, ³Gliwice/PL
Purpose: The aim of the study is to examine the correlation between DTI findings, clinical motor and sensory deficits, and motor evoked potential (MEP) parameters. This will provide insights into the potential of DTI metrics as biomarkers for predicting functional recovery and guide therapeutic interventions in patients with chronic spinal cord injury (SCI).
Methods or Background: A total of 29 patients with SCI (both paraplegic p-SCI and tetraplegic t-SCI), matched by sex and age to 29 healthy controls, were neurologically and neurophysiologically evaluated, including MEPs recorded from upper and lower limb muscles. Diffusion tensor imaging (DTI) was performed using a 3 Tesla MRI scanner and processed using Human Motor Area (HMAT) and Sensorimotor Area Tract (SMATT) templates.
Results or Findings: No significant DTI differences were found between p-SCI and t-SCI or p-SCI and healthy controls. However, patients with t-SCI had lower fractional anisotropy (FA) in primary motor (M1) and sensorimotor (S1) tracts, pre-supplementary motor area (pre-SMA) tracts, M1 and S1 cortices, and left pre-SMA cortex compared to controls. In t-SCI patients, higher motor scores correlated with increased FA in ventral premotor area (PMv) tracts and cortices, and higher sensory scores with higher FA in S1 tracts. MEP amplitudes from rectus femoris also positively correlated with FA in motor tracts, M1, PMd, PMv, and SMA cortices.
Conclusion: DTI findings reveal distant degeneration in the sensorimotor cortex and supraspinal tracts in chronic SCI, which correlates with clinical motor and sensory scores, as well as MEP parameters from rectus femoris muscles in t-SCI patients. DTI metrics can serve as potential biomarkers to predict motor and sensory recovery in patients with SCI and to guide and track therapeutic interventions.
Limitations: Cross-sectional design and the small sample size.
Funding for this study: Grant NCBiR ERA-NET-NEURON/13/2018, Wroclaw Medical University grant SB.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study was performed in accordance with the Declaration of Helsinki and was approved by the Bioethics Committee of the Wroclaw Medical University.

Author Block: J. M. E. Jende¹, C. Mooshage¹, K. Zhang¹, T. Platt¹, C. Neelsen¹, M. Bendszus¹, H-P. Schlemmer¹, M. Ladd¹, F. Kurz²; ¹Heidelberg/DE, ²Geneva/CH
Purpose: Disorders of the peripheral nervous system such as polyneuropathies pose a huge challenge to the global healthcare system. The exact pathophysiology underlying most polyneuropathies remains poorly understood. Previous studies on 3T MR neurography (MRN) have found that the maximum of fascicular nerve damage in various polyneuropathies is located at the level of the sciatic nerve although clinical symptoms usually occur further distally.
Methods or Background: To understand the clinical impact and physiological background of fascicular sciatic nerve lesions, 10 patients with distal symmetric polyneuropathy and 10 healthy controls matched for age and BMI underwent T2-weighted, high resolution MRN of the right thigh at 3T and 7T. At 7T, the maximum fascicular diameter and the average number of nerve fascicles per slice were measured. At 3T, additional diffusion-weighted and T2-relaxometry sequences were acquired and the sciatic nerve’s fractional anisotropy (FA) and T2 relaxation times (T2R) were calculated.
Results or Findings: At 7T, the sciatic nerve’s maximum fascicular diameter in patients with polyneuropathies was larger compared to controls (1.41mm±0.16 vs. 1.03mm±0.07; p=0.049mm). In patients with polyneuropathies, the fascicular diameter was negatively correlated with the number of nerve fascicles (r=-0.72;p=0.018) and the FA (r=-0.78;p=0.017mm). Positive correlations were found between the fascicular diameter and T2R (r=0.76;p=0.019).
Conclusion: The results indicate that polyneuropathies cause a fusion of nerve fascicles that results in a reduced number of nerve fascicles and a larger fascicular diameter that is associated with a structural and functional decline represented by a decrease in nerve FA and an increase in T2R. This study is the first to show that the fascicular diameter of the sciatic nerve is directly related to changes in FA, T2R and clinical neuropathy status. The causes of fascicular fusion remain to be determined.
Limitations: Cohort Size
Funding for this study: Else Kröner Fresenius Foundation (EKFS)
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was appoved by the local ethics committee of Heidelberg University Hospital.

Author Block: F. Ensle¹, F. Zecca², B. J. Kerber¹, M. Lohezic¹, J. Kroschke¹, K. Pawlus¹, R. Guggenberger³; ¹Zurich/CH, ²Cagliari/IT, ³Winterthur/CH
Purpose: To assess and compare DESS and post-contrast STIR sequences in deep-learning(DL)-reconstructed 3D MR neurography of the extraforaminal cranial and spinal nerves.
Methods or Background: Eighteen consecutive exams of 18 patients with unclear cephalgia undergoing head-and-neck MRI at 1.5T were retrospectively included (mean age: 51 ± 14 years, 11 female). 3D DESS and post-contrast 3D STIR sequences were reconstructed with a prototype DL algorithm. Two blinded readers qualitatively evaluated visualization of the inferior alveolar (IAN), lingual (LN), facial (FN), hypoglossal (HN), greater occipital (GON), lesser occipital (LON) and greater auricular (GAN) nerves, as well as overall image quality, vascular suppression and artifacts. Apparent signal-to-noise ratio (aSNR) and contrast-to-noise ratios (aCNR) were measured. Qualitative ratings were compared between sequences using Wilcoxon signed-rank test, quantitative analysis with paired sample Student’s t-testing.
Results or Findings: DESS demonstrated significantly improved visualization of the LON and GAN and proximal GON (p < 0.015). Post-contrast STIR showed significantly enhanced visualization of the LN, HN and distal IAN (p < 0.001). The FN, proximal IAN and distal GON did not demonstrate significant differences in visualization between DESS and post-contrast STIR (p > 0.08).
With regard to overall image quality and artifacts, there was also no significant difference between sequences. Post-contrast STIR achieved superior vascular suppression, reaching statistical signifi-cance for one reader (p = 0.039). Quantitatively, there was no significant difference between sequences (p > 0.05).
Conclusion: Our findings suggest that 3D DESS generally provides improved visualization of spinal nerves (GON, LON, GAN), while post-contrast 3D STIR facilitates enhanced delineation of extraforaminal cranial nerves (IAN, LN, HN). 3D DESS and post-contrast 3D STIR could each add value to head-neck MRN protocols, depending on the main clinical area of interest.
Limitations: Retrospective study. Limited sample size.
Funding for this study: Not applicable.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the institutional review board.

Author Block: V. Sedlák, K. Vambersky, A. Kavková, K. Sichova, D. Netuka, T. Belsan, M. Majovsky; Prague/CZ
Purpose: The objective of this study is to demonstrate how the application of advanced denoising techniques, such as MP-PCA and P2S, along with post-processing methods for enhancing angular resolution, can significantly reduce imaging times in advanced diffusion imaging. This approach aims to preserve or improve image quality while reducing the acquisition time burden in clinical and research settings.
Methods or Background: We acquired two sets of advanced diffusion MRI data from patients with glial brain tumors: one using a full-length acquisition and the other using a fast acquisition protocol. The full-length dataset was processed directly, while the fast protocol data was enhanced with denoising algorithms (e.g. MP-PCA, P2S) and angular super-resolution reconstruction techniques. We evaluated data quality by assessing signal-to-noise ratio (SNR), angular resolution, and diagnostic accuracy for predicting glioma grade and IDH mutation status.
Results or Findings: Data from 100 patients with glial brain tumors were processed and analyzed. The shortened protocol data, when enhanced by MP-PCA and P2S denoising, provided a significant improvement in SNR, aligning closely with or even superseding the quality of the full-length acquisition. Angular super-resolution reconstruction techniques further enhanced the angular resolution without extending scan time. The diagnostic accuracy for predicting glioma grade and IDH mutation status was comparable between the full-length and processed shortened datasets, demonstrating that these post-processing methods can preserve clinical diagnostic value while reducing scan duration
Conclusion: Denoising techniques and angular resolution enhancement significantly improve the quality of shortened diffusion MRI acquisitions, maintaining diagnostic accuracy for glioma grading and IDH status while reducing scan time.
Limitations: Results are based on 100 glioma patients and require validation in broader populations. The computational demands may limit immediate clinical application, and performance may vary with MRI hardware and sequence parameters.
Funding for this study: This study was supported by the Grant Agency of Charles University, grant number GAUK 222623
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Approved by the Ethics committee of the Military. University Hospital Prague

Author Block: X. Fan, G. Cheng, X. Zhang, N. Zhang; Shenzhen/CN
Purpose: Our objectives are to depict the whole white matter DTI-ALPS changes pattern in cognitive impairment and provide an intuitive and comprehensive method to assess the activity of the glymphatic system.
Methods or Background: The glymphatic system is increasingly recognized as a critical factor in the pathogenesis of dementia. We creatively propose a novel method to analyze the whole white matter diffusion tensor image analysis along the perivascular space (DTI-ALPS). We included 304 participants from the Shenzhen Multimodal Aging Research (STAR) Cohort recruited in Peking University Shenzhen Hospital, including 182 cognitively unimpaired (CU) participants, 93 participants with mild cognitive impairment, and 29 patients with dementia. All participants underwent 3.0T MRI scans with DTI sequences. We calculated the DTI-ALPS values using the conventional method and depicted the whole white matter DTI-ALPS mapping with the novel method we developed. We performed the analysis of variance (ANOVA) to find the differences in ALPS values among three groups and used post-hoc tests to find intergroup differences in the conventional method and DTI-ALPS mapping, respectively.
Results or Findings: We revealed the whole brain ALPS pattern using the DTI-ALPS mapping. Over 13 out of 30 regions showed significant differences (p < 0.05) in the inter-group analysis, which provided additional information beyond the significant differences based on the conventional ROI-based ALPS.
Conclusion: In conclusion, the DTI-ALPS mapping provides a robust, intuitive, and comprehensive way to evaluate the changing pattern of the glymphatic system, overcoming the limitations introduced by conventional ROI-based DTI-ALPS calculating methods.
Limitations: Future studies should integrate additional statistics such as minimum, maximum, and standard deviation into ALPS-mapping analyses for a comprehensive understanding of the DTI-ALPS mapping.
Funding for this study: This study is principally supported by the Shenzhen Science and Technology Program (KCXFZ 20211020163408012).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: the Ethics Committee of Peking University Hospital

Author Block: S. J. Ahn, W. J. Son, J. Y. Lee, H. Lee; Seoul/KR
Purpose: Changes of brain hippocampal volumes are closely associated with the development of Alzheimer’s disease. In this work, we develop a new deep learning (DL) network model for volumetric hippocampal segmentation from computed tomography (CT) head images, a task that has been challenged due to the modality’s limited brain contrast.
Methods or Background: HDD-Net: The proposed network model is characterized by four major elements – 1) an encoder , 2) two parallel decoders (namely, seg-decoder and edge-decoder), 3) a feature cross module (FCM) fusing features from the two decoders and 4) a cross loss computing differences between outputs .

Datasets and preprocessing: 150 pairs of MRI-CT volumetric head images collected at Gangnam Severance Hospital were used for model training (N=120) and internal validation (N=30), while 47 pairs were selected from Seoul St. Mary’s Hospital database for external validation. Ground-truth hippocampal labels were generated from T1-weighted MR images using FreeSurfer, and then were processed by a Gaussian high-pass filter leading to reference edge maps. Each pair of MR-CT images were coregistered using SPM12.

Training and Evaluation: The DL model was trained with a cost function combining segmentation, edge, and cross losses. Its performance was evaluated by calculating Dice coefficient and intersection-over-union (IOU). The performance our model was compared with that of conventional U-Net model.
Results or Findings: Dice and IOU of our model is higher than those of U-net for internal validation set (DICE : 0.840 vs. 0.822, IOU: 0.726 vs. 0.699) and external validation (DICE : 0.784 vs. 0.751, IOU: 0.650 vs. 0.613)
Conclusion: Results suggest feasibility of DL-based automated hippocampal segmentation from CT scans and its improved performance via edge decoding.
Limitations: The applicability of the model could be enhanced by training it on patients with dementia
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Gangnam severance hospital IRB

Author Block: Y. Han¹, L. Peng², T. Meng², Q. Sun¹, G. Zhang², T. Wang², X. Wang¹; ¹Jinan/CN, ²Shanghai/CN
Purpose: To describe the initial experience and evaluate the clinical feasibility of 60kVp craniocervical CT angiography (CTA) with artificial intelligence iterative reconstruction (AIIR).
Methods or Background: Sixty consecutive patients scheduled for craniocervical CTA were prospectively enrolled and underwent a 60kVp (350 ref. mAs) and a followed 120kVp (120 ref. mAs) craniocervical CTA, in a 5-minute interval, using two separate contrast medium (CM) injections. The 120kVp scans were reconstructed with hybrid iterative reconstruction (HIR), while the 60kVp scans were reconstructed with HIR and AIIR. Two radiologists diagnosed the stenosis, the intracranial aneurysm (IA), and the vascular anatomic variant (VAV) in consensus using a 5-point confidence scale (1=definitely absent, 5=definitely present) on a per-patient basis, which was used for a receiver operating characteristic analysis, and evaluated the vessel visibility (1=blur, 5=clear). Image noise on the common carotid artery (CCA) was measured. The diagnostic performance and image quality of 60kVp scans were evaluated, using 120kVp scans as the reference standard.
Results or Findings: The mean effective dose and CM dosage was 0.18 ± 0.04 mSv and 30.58 ± 5.15 ml, respectively, for 60kVp acquisition, corresponding to an 83.02% and 38.46% reduction as compared to 120kVp acquisition. Under 60kVp acquisition, AIIR outperformed HIR in diagnosing all three abnormal manifestations, showing higher AUC (stenosis: 0.98 vs 0.57; IA: 0.92 vs 0.62; VAV: 0.92 vs 0.79; all p<0.05). No significant difference in vessel visibility was found between 60kVp AIIR and reference images (4.32±0.97 vs 4.42±0.72, p=0.412), while AIIR images showed lower image noise (10.14±4.64 HU vs 14.68±5.23 HU; p<0.05).
Conclusion: The 60kVp craniocervical CTA with AIIR has the potential for profound dose reduction without compromising the image quality and the diagnostic performance.
Limitations: N/A
Funding for this study: N/A
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the local IRB