RPS 411 - Aging brain and neurodegeneration imaging

Author Block: M. M. Courtney, R. A. Kenny, J. F. Meaney, C. De Looze; Dublin/IE
Purpose: White matter hyperintensities (WMHs) are known to correlate with cognitive decline, stroke, and dementia. Previous research has explored the independent effects of white matter macrostructure, microstructure, and spatial distribution on cognitive function, yet a comprehensive analysis combining elements is limited. We provide a comprehensive WMH analysis to assess their association with cognitive decline over a six-year period.
Methods or Background: Data was obtained from The Irish Longitudinal Study on Ageing (TILDA), including MRI scans and cognitive performance scores from 497 community-dwelling older adults. WMHs were segmented using Lesion Prediction Algorithm, analysed using Explore DTI for diffusion metrics. Statistical Analysis performed in R-Studio. Linear mixed effect models used to assess relationship between lesion phenotypes and cognitive decline, adjusting for demographic and health-related variables.
Results or Findings: 11,933 WMHs analysed. Average 24 lesions per subject. Average lesion volume 263mm3. Average lesion FA 0.29, MD 1.10. K-means clustering identified 3 primary WMH phenotypes. Deep WMHs associated with older age and 2 or more cardiovascular risk factors (p<0.001 respectively). Higher volume lesions were associated with cardiovascular risk factors (p<0.001), smoking (p<0.01) and older age (p<0.001). High-volume, low FA lesions in both deep (p=0.5) and periventricular (p=0.04) white matter exhibited accelerated cognitive decline over six years. Increased number of periventricular lesions was associated with cognitive decline (p<0.01) .
Conclusion: WMHs manifest diverse phenotypes associated with cognitive decline. High-volume, low FA lesions in both periventricular and deep white matter are predictive of cognitive deterioration. Identifying WMH phenotypes may inform early intervention strategies and improve patient outcomes by targeting individuals at higher risk of cognitive decline.
Limitations: Cross-sectional MRI data implies findings are associative and correlate with longitudinal findings, but correlation does not equal causation. Further investigation with serial MRI would provide more reliable data.
Funding for this study: The Irish Longitudinal Study on Ageing is funded by the Irish Department of Health, the Atlantic Philanthropies and Irish Life.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the Trinity College Faculty of Health Sciences Research Ethics Committee, Dublin, Ireland. Protocols conformed with the Declaration of Helsinki. Signed informed consent was obtained from all respondents prior to participation. Additional ethics approval was received for the magnetic resonance imaging (MRI) sub-study from the St James’s Hospital/Adelaide and Meath Hospital, Inc. National Children’s Hospital, Tallaght (SJH/AMNCH) Research Ethic Committee, Dublin, Ireland. Those attending for MRI also completed an additional MRI-specific consent form. (De Looze et al)

Author Block: L. Yuna; Beijing/CN
Purpose: Disentangling the complex interaction between aging and various aging-related neurological diseases at the brain network level.
Methods or Background: We presented a unified analysis framework to identify aging-related morphological connectivity networks (MCNs) and determined their clinical relevance in various neurological diseases (including mild cognitive impairment, Alzheimer's disease, Parkinson’s disease, small vessel disease multiple sclerosis and multiple sclerosis). First, individual MCNs in the HC group were constructed and further decomposed into distinct subnetworks using linked independent component analysis. Aging-related subnetworks were defined as those significantly associated with age. The aging-related subnetworks were spatially correlated with disease-related MCN disruptions. Further, the regression coefficients of the aging-related subnetworks were calculated for each patient’s MCN using linear regression. The regression coefficients were then correlated with various clinical variables within each disease group to assess the clinical significance of the aging-related subnetworks. Finally, a series of annotated biological maps were utilized to advance the biological interpretation of the identified aging-related subnetworks.
Results or Findings: We first identified three aging-related subnetworks, including the perceptual-limbic subnetwork, attention-somatomotor subnetwork, and somatomotor-predominant subnetwork, that exhibited distinct aging trajectories. Normal aging interacted with various neurological diseases, exhibiting both transdiagnostic and diagnosis-specific patterns at the brain network level. The aging-related subnetworks were closely related to cognitive and physical performance in patients. Biological correlation analysis revealed that glucose metabolism and several neurotransmitters, such as cannabinoids and dopamine, played critical roles in aging-related subnetworks.
Conclusion: This study elucidated the network mechanisms underlying the complex interactions between aging and neurological diseases, offering insights that could improve clinical management and therapy development by evaluating aging effects.
Limitations: This study is limited by an uneven sample distribution, variability in disease durations, and the absence of longitudinal research
Funding for this study: We demonstrated how normal aging interacted with various neurological diseases at the brain network level, both transdiagnostically and diagnosis-specifically. The identified aging-related subnetworks might serve as imaging markers to distinguish normal aging effects from disease-specific mechanisms, thereby improving disease monitoring and management.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Beijing Tiantan Hospital, Capital Medical University, Beijing, China; No. KY-2019-050-02

Author Block: G. Di Cerbo, G. Saltarelli, A. Innocenzi, M. Cella, C. De Felici, F. Bruno, A. Splendiani, E. Di Cesare; L'Aquila/IT
Purpose: The purpose of this study is to compare the operating features and analysis outputs of four different commercially available software for brain volumetric analysis.
Methods or Background: We analyzed consecutive brain MRI scan of 32 patients (25 males, aged between 50 and 90 years) evaluated in a singles Institution for cognitive decline. All MRI examinations were performed on 3T scanner (GE MR750w.), including a volumetric T1 GRE sequence (slice 1 mm, TR 8.5, frequency FoV 25.6, phase FoV 0.8).
MRI data were analyzed through four different dedicated softwares (S1, S2, S3, S4) after quality check by an experienced neuroradiologist.
Volumetric output data of brain segmentation and volume for frontal, temporal, parietal, occipital lobes, hippocampus, and lateral ventricles, were collected and compared.
Results or Findings: The results revealed no significant consensus among the four artificial intelligence software applications in measuring various brain areas.
S1-S2 showed non statistically significant output values in all brain regions.
S1-S3 showed statistically significant differences in frontal and parietal lobe, lateral ventricles and hippocampus.
S1-S4 showed statistically significant differences in frontal parietal and occipital lobe, and lateral ventricles.
S2-S3 showed statistically significant differences in temporal and occipital lobes.
S3-S4 showed statistically significant differences in parietal and occipital lobes, hippocampus and lateral ventricles.
Conclusion: Although AI software are becoming increasingly popular in clinical practice, the findings indicate a low degree of concordance among the four applications evaluated in this study. Therefore, clinicians integrating these tools into routine practice should be aware of the limited result interchangeability across different software platforms and consider their use as complementary aids rather than substitutes for clinical expertise.
Limitations: Small sample size
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Local IRB

Author Block: X. Xu, B. Zhang, Z. Zhu; Nanjing/CN
Purpose: To elucidate the pathological mechanism of glymphatic system dysfunction are associated with regional tau deposition and tau-mediated neurodegeneration across the preclinical and prodromal stage of the AD continuum.
Methods or Background: Cognitively normal (CN) controls (n=94), individuals with mild cognitive impairment (MCI; n = 83), and those with significant memory concern (SMC; n =84) were included from the Alzheimer's Disease Neuroimaging Initiative. Tau pathology was measured by positron emission tomography, the glymphatic activity assessed by diffusion tensor image analysis along the perivascular space (DTI-ALPS), and neurodegeneration reflected by hippocampal volume. Mediation analysis was used to study the possible pathways.
Results or Findings: ALPS was significantly associated with tau and tau-mediated neurodegeneration, especially in parahippocampal gyrus. The relationship between glymphatic function and neurodegeneration was mediated by tau pathological deposition (indirect effect: 0.012, 95%CI [0.001—0.029]) rather than ALPS index mediated the relationship between tau and neurodegeneration (indirect effect: 0.212, 95%CI [-0.013,0.0002]). The relationship between glymphatic dysfunction and cognitive decline were fully mediated by tau deposition and neurodegeneration in preclinical AD.
Conclusion: Tau deposition in specific region may mediate the relationship of glymphatic dysfunction and neurodegeneration, which contribute to cognitive decline in the preclinic AD stage, facilitating the development of therapeutics targeting tau protein and glymphatic dysfunction in AD .
Limitations: Firstly, the cross-sectional design employed in this study limits us to explore the causal relationships or investigate longitudinal changes over time, future longitudinal studies could provide further. Secondly, the ALPS index is mainly used to measure the function of the subcortical glymphatic system. Future research should continue to use other methods, such as BOLD-CSF coupling measurements, to validate the function of the subcortical glymphatic system in AD.
Funding for this study: This work was supported by the National Science and Technology Innovation 2030 -- Major program of "Brain Science and Brain-Like Research" (2022ZD0211800); the National Natural Science Foundation of China (82271965, 81971596, 82001793); the Fundamental Research Funds for the Central Universities, Nanjing University (2020-021414380462); the Key Scientific Research Project of Jiangsu Health Committee (K2019025); Special Funded Project of Nanjing Drum Tower Hospital (No. RC2022-023), Development Plan (Social Development) Project of Jiangsu Province (No. BE2022679). China Postdoctoral Science Foundation (2023M741648). The National Natural Science Foundation of China (82302172); The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: ADNI Ethics committee

Author Block: B. Atalay, K. Erincik, M. B. Doğan, M. Gezgin, H. Yıldız, F. B. Ozdilek; Istanbul/TR
Purpose: To assess the impact of temporal muscle thickness (TMT), as an indicator of sarcopenia on cognitive status and medication dosage in patients with Parkinson's disease (PD).
Methods or Background: A total of 54 patients with PD and 46 healthy controls were retrospectively analyzed. Brain MR images from both groups were reviewed by two radiologists who independently measured the right and left temporal muscle thickness using T1-weighted axial images. Clinical assessments included the Unified Parkinson’s Disease Rating Scale (UPDRS), Mini-Mental State Examination (MMSE), Hoehn and Yahr Scale, L-dopa equivalent daily dose (LEDD), and disease duration, collected by a neurologist. Interobserver agreement was evaluated using the intraclass correlation coefficient. The relationship between TMT and clinical data was analyzed using Spearman’s correlation.
Results or Findings: In the PD group, 33.3% of patients were female, compared to 50% in the control group. Interobserver agreement for TMT measurements was excellent. No significant difference in TMT was observed between the PD and control groups (p=0.16, p=0.34). A weak but statistically significant correlation was found between TMT, LEDD, and disease duration, while no correlation was found between TMT and UPDRS or Hoehn and Yahr scores. A weak but significant correlation was observed between TMT and MMSE scores.
Conclusion: Sarcopenia, characterized by muscle mass loss, shares contributing factors with Parkinson’s disease. TMT, a reliable marker of muscle mass, can be measured on brain MRIs to assess sarcopenia risk in PD patients. Our findings suggest that TMT correlates with cognitive function and medication dosage in PD patients, making it a valuable tool for early sarcopenia detection and management in clinical settings.
Limitations: Lack of patient follow-up and inability to assess the impact of muscle mass increase on medication dosage.
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: The ethics committee notification can be found under the number 2023/0912.

Author Block: S. Rajan, J. S. Chatha, H. Mahajan; New Delhi/IN
Purpose: To assess whether reconstruction of phase images from susceptibility-weighted imaging (SWI) in a coronal plane enhances the visualization confidence of the nigrosome-1 layer in the substantia nigra. The coronal Swallow Tail cleft sign may serve as a valuable indicator for this structure.
Methods or Background: A retrospective review was conducted on 433 consecutive MR brain scans acquired using a routine protocol for various indications. The axial phase images of SWI were analyzed in Phase 1 by two radiologists with 9 and 22 years of experience. They evaluated the visibility of the nigrosome-1 layer below the level of the red nucleus using a 5-point Likert scale (1 = very difficult to 5 = easily seen). In Phase 2, coronal reconstructed images of the phase SWI were assessed perpendicularly to the substantia nigra in posterior sections for the presence of the cleft sign, also scored with the 5-point Likert scale.
Results or Findings: In Phase 1, the Likert scores of 1 and 2 were similar (6% and 9%, respectively). There was a marginal decrease in score 3 (from 4.97% to 4.04%). Scores of 4 decreased significantly from 16.97% in Phase 1 to 7.96% in Phase 2, while the score of 5 increased substantially from 62.01% in Phase 1 to 72.97% in Phase 2. Statistical analysis demonstrated that the addition of the coronal Swallow Tail cleft sign significantly enhanced confidence in visualizing the nigrosome-1 layer (p < 0.01).
Conclusion: The coronal Swallow Tail cleft sign significantly improves the confidence of visualization of the nigrosome-1 layer in both normal subjects and those with Parkinsonism. This finding underscores the utility of coronal reconstructions in enhancing diagnostic accuracy in neuroimaging.
Limitations: Correlation with nuclear scans or clinical history was not done
Funding for this study: None
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: Restrospective

Author Block: W. Cheng; Nanchang/CN
Purpose: Parkinson’s disease (PD) and progressive supranuclear palsy (PSP) present similar clinical symptoms, but their treatment options and clinical prognosis differ significantly. Therefore, we aimed to discriminate between PD and PSP based on multi-level indices of rs-fMRI via the machine learning approach.
Methods or Background: A total of 58 PD and 52 PSP patients were prospectively enrolled in this study. Participants were randomly allocated to a training set and a validation set in a 7:3 ratio. Various resting-state functional magnetic resonance imaging (rs-fMRI) indices were extracted, followed by a comprehensive feature screening for each index. We constructed fifteen distinct combinations of indices and selected four machine learning algorithms for model development. Subsequently, different validation templates were employed to assess the classification results and investigate the relationship between the most significant features and clinical assessment scales.
Results or Findings: The classification performance of logistic regression (LR) and support vector machine (SVM) models, based on multiple index combinations, was significantly superior to that of other machine learning models and combinations when utilizing automatic anatomical labeling (AAL) templates. This has been verified across different templates.
Conclusion: The utilization of multiple rs-fMRI indices significantly enhances the performance of machine learning models and can effectively achieve the automatic identification of PD and PSP at the individual level.
Limitations: Only the rs-fMRI index was used in this study, and DTI-related microstructure data was not included.
Funding for this study: This study was supported by the National Natural Science Foundation of China (82160331), Jiangxi Province Double Thousand Talent Plan (jxsq2023201039). This project is implemented by the Jiangxi Clinical Research Center for Medical Imaging (20223BCG74001), and Jiangxi Province Key Laboratory for Precision Pathology and Intelligent Diagnosis (2024SSY06281).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This study was approved by the Medical Ethics Committee of the First Affiliated Hospital of Nanchang University (approval number: IIT2022124).

Author Block: M. Sarkinaite¹, U. Lukoseviciute¹, N. Masiulis¹, S. Lukoševičius¹, O. Levin², R. Gleiznienė¹; ¹Kaunas/LT, ²Leuven/BE
Purpose: Subcutaneous fat accumulation has been linked to adverse brain health outcomes. This study examines the relationship between thigh muscle mass, subcutaneous fat distribution, and brain structure in elderly adults. It explores how body composition affects cortical thickness in brain regions linked to cognitive function.
Methods or Background: Fifty-four healthy elderly individuals underwent imaging of the right thigh and brain using a 3T Siemens Avanto MRI system. Muscle and subcutaneous fat cross-sectional areas (CSA) were measured at 50% and 20% of thigh length, with the muscle-to-fat ratio calculated at the 50% mark. Cortical thickness was assessed through brain volumetric analysis using Freesurfer 7.4.1 software.
Results or Findings: Significant positive correlations (r ≥ 0.2, p ≤ 0.05) were found between the muscle-to-fat ratio and cortical thickness in the left cerebellum, cuneus, and transverse temporal cortex, in the right entorhinal cortex, inferior temporal cortex, postcentral gyrus, superior temporal cortex, and the banks of the superior temporal sulcus (BANKSSTS). Additionally, significant negative correlations (r ≤ -0.2, p ≤ 0.05) were observed between subcutaneous fat CSA at 50% of thigh length and cortical thickness in the left cuneus and entorhinal cortex, in the right BANKSSTS, postcentral gyrus, and superior temporal cortex. Additionally, subcutaneous fat CSA at 20% of thigh length was inversely correlated with cortical thickness in the left cuneus and right BANKSSTS (r ≤ -0.2, p ≤ 0.05).
Conclusion: Our findings demonstrate that increased thigh muscle mass correlates with greater cortical thickness in cognitive regions, while elevated subcutaneous fat is linked to reduced thickness. These results highlight the role of body composition in maintaining brain health in the elderly and underscore the importance of muscle mass in mitigating age-related cortical decline.
Limitations: With only 54 participants, the study's findings may lack generalizability.
Funding for this study: Supported by the Research Council of Lithuania (grant number P-MIP-22-217).
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Approved by the Kaunas region Medical Ethics Committee for Biomedical Research (No. BE-2-22).

Author Block: B. Genç, I. C. Koc, K. Aslan; Samsun/TR
Purpose: Peak skeletonized mean diffusivity (PSMD) is being proposed as a novel biomarker for small vessel disease. Tract-Based Spatial Statistics (TBSS) is a well-established DTI analysis method that enables the fully automated detection of microstructural changes in white matter. The aim of this study is to investigate white matter changes in patients with age-related hearing loss using both PSMD and TBSS.
Methods or Background: All individuals from the OpenNeuro hearing loss connectome dataset were included in the study (https://openneuro.org/datasets/ds005026). The dataset consisted of 52 hearing loss (HL) patients and 30 healthy controls. From the DTI images in the dataset, FA, MD, RD, and AD were obtained using FSL with preprocessing steps including TOPUP and eddy. The standard TBSS procedure was applied to investigate group differences in FA, MD, RD, and AD. The standard PSMD method with histogram analysis was used to calculate the difference between the 95th and 5th percentile MD values (https://www.psmd-marker.com/). A comparison between the groups was made.
Results or Findings: The TBSS analysis did not show any differences between the groups for any of the DTI parameters. The PSMD value in the HL group was 225,21±23,50 x 10^-6 mm²/s, while in the control group it was 214,53±24,01 x 10^-6 mm²/s, showing a statistically significant increase in the hearing loss group (p=0,039).
Conclusion: Our findings suggest that small vessel disease may underlie the pathophysiology in patients with age-related hearing loss. To our knowledge this is the first study to show the relationship between age-related hearing loss and small vessel disease.
Limitations: Since numerical data of hearing tests were not available, the correlation between PSMD data and hearing test could not be analyzed.
Funding for this study: No funding was received for this study.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: OpenNeuro permits the use of patient data under the CC0 license. Ethical approval has already been obtained by the "University of Salerno." Therefore, no separate ethical approval has been obtained from our institution.

Author Block: F. Chen¹, B. Zhang¹, Q. Chen¹, X. Fan², L. Zou², Y. Li², G. Cheng², G. Danni¹; ¹Nanjing/CN, ²Shenzhen/CN
Purpose: To assess the feasibility of analysis along the perivascular space (ALPS) using DTI obtained from 5.0 Tesla MR, assess glymphatic system function in SCD and its correlation with spatial navigation abilities and plasma biomarkers.
Methods or Background: Glymphatic dysfunction is implicated in cognitive impairment associated with AD. Spatial navigation impairments are among the earliest manifestations in individuals with SCD. However, the relationship between glymphatic function and spatial navigation remains poorly understood. Between May 2023 and January 2024, 62 SCD patients and 62 matched controls underwent high-resolution DTI on 5.0T MR scanner, spatial navigation behavioral tests, cognitive assessments, and Simoa plasma biomarker analyses. The ALPS index reflecting glymphatic activity was calculated by a ratio of the diffusivities along the x-axis in the projection and association neural fibers to the diffusivities perpendicular to them and compared according to the groups with use of multivariate analysis of variance. Inter-reproducibility of ALPS index among 5.0TMR and 3.0TMR scanners was evaluated using consistency interclass correlation coefficient. Pearson correlation analysis was used to assess the relationship between cognitive performance, spatial navigation performance, plasma biomarkers, and the ALPS index.
Results or Findings: The ALPS on 5.0TMR and ALPS on 3.0TMR showed strong consistency and correlation. SCD patients had significantly lower ALPS index on 5.0TMR and higher navigation errors compared to controls. The ALPS index was positively correlated with spatial navigation, cognitive performance, and memory performance, and negatively correlated with plasma pTau217 levels.
Conclusion: The glymphatic function is impaired in SCD at the preclinical AD stage, which may represent one of the physiological mechanisms leading to deficits in spatial navigation abilities. DTI-ALPS on 5.0T MR may serve as a sensitive neuroimaging biomarker for the preclinical stage of AD.
Limitations: Our study is a small-sample cross-sectional study.
Funding for this study: National Science and Technology Innovation 2030 -- Major program of "Brain Science and Brain-Like Research" (2022ZD0211800)
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The Research Ethics Committees of Nanjing Drum Tower Hospital, the Shenzhen Institute of Advanced Technology of the Chinese Academy of Sciences, and Peking University Shenzhen Hospital.

Author Block: G. Danni, Z. Bing; Nanjing/CN
Purpose: To investigate the static functional connectivity (sFC) and dynamic functional connectivity (dFC) of medial entorhinal cortex (MEC) and lateral entorhinal cortex (LEC) in individuals with subjective cognitive decline (SCD) and the associations with cognitive performance, spatial navigation and olfactory memory.
Methods or Background: Seventy-seven control subjects and 106 SCD individuals were enrolled, and neuropsychological evaluations, 2D computerized spatial navigation test, olfactory memory test and resting-state functional magnetic resonance imaging (rs-fMRI) were collected. Bilateral MEC and LEC were selected as seeds to investigate alternations of the volumes, sFC and dFC.
Results or Findings: Compared to control subjects, SCD individuals exhibited decreased sFC between bilateral LEC and visual network, between right LEC and left posterior cingulate gyrus and sensory motor network, and between right MEC and left hippocampus, visual network and sensory motor network. The dFC between right LEC and right triangular part of inferior frontal gyrus (IFGtriang) decreased, while dFC between left MEC and right putamen, and between right MEC and right middle temporal gyrus increased. In SCD group, volumes of bilateral MEC were positively correlated with spatial navigation ability, and sFC between bilateral LEC and visual network was positively correlated with olfactory memory. The dFC between right LEC and right IFGtriang was correlated positively with global cognitive performance. The combination of sFC and dFC as biomarkers to identify SCD showed an area under curve of 92.1%.
Conclusion: There were functional alternations of EC subregions in SCD individuals, and we demonstrated the association between LEC and spatial navigation, and MEC and olfactory memory. The combination of sFC and dFC may be a new neuroimaging biomarker for the early diagnosis of AD.
Limitations: The study lacked genetic and biomarker data. Also, we didn't have follow-up data to track pathological progression.
Funding for this study: National Science and Technology Innovation 2030_Major program of "Brain Science and Brain-Like Research" (No. 2022ZD0211800)
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Nanjing Drum Tower Hospital Ethics Committee

Author Block: S. Van Eyndhoven, R. Magalhaes, R. Khan, T. V. Phan, A. Liseune, A. Brys, D. M. Sima, J. Verheyden, A. Ribbens; Leuven/BE
Purpose: Develop a robust automated deep learning-based method for assessment of cerebral microhemorrhages on T2* gradient-echo (GRE) images.
Methods or Background: Hypointensities on GRE images can be indicative of cerebral microhemorrhages, and serve as exclusionary criteria for anti-amyloid therapies for Alzheimer’s disease, as they are linked to increased risk of intracerebral hemorrhage. Automated detection tools could greatly aid radiologists in the challenging task of accurately quantifying these findings. A deep learning model was developed to detect microhemorrhages on cross-sectional GRE images.

Training was done using 247 2D GRE images and accompanying manual microhemorrhage annotations from the EMERGE clinical trial (NCT02484547). Detection accuracy was evaluated on a stratified subset of cases (N=600) of the Alzheimer’s Disease Neuroimaging Initiative, where up to 10 microhemorrhages were annotated on a 2D GRE sequence by experts. This is representative of the population that would be screened via MRI before administration of anti-amyloid treatment, for which assessment is more challenging relative to cases with a large number of microhemorrhages.

For each case, we evaluated the F1 score, i.e., the harmonic mean between detection sensitivity and positive predictive value, and the absolute error between the number of microhemorrhages according to the expert ground truth and the automated count.
Results or Findings: The median F1 score of the trained model was 0.67, and its median absolute count error was 1 microhemorrhage.
Conclusion: Detection of microhemorrhages on 2D GRE images is a challenging task, which is gaining importance with the advent of novel anti-amyloid treatments that may lead to hemorrhagic side effects. Using a robust, validated AI tool, as described here, can assist radiologists in microhemorrhage detection, providing value especially in sparse MR images, though expert assessment remains vital.
Limitations: None.
Funding for this study: N/A
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: N/A