Clinical follow-up assessment using chest X-rays (CXRs) requires radiologists to interpret both individual images and, crucially, the temporal progression of pathologies between baseline and follow-up examinations. To address this challenging task of follow-up medical visual question answering (MVQA), we proposed DoKE, domain knowledge-enhanced multimodal framework, which integrated two core innovations: (1) an intra-and inter-image extraction module (IIIE), which employed a two-stream architecture with intra-and inter-image self-attention to first enhance salient regions within each CXR and then explicitly model semantic correspondences and progression-related differences across the image pair; and (2) a domain knowledge enhancement scheme (DKE) that injected structured diagnostic prior, disease probabilities from a classifier covering tail-end pathologies, to guide the visual representation learning. Extensive experiments on the MIMIC-Diff-VQA benchmark demonstrated that DoKE outperformed prior state-of-the-art methods across key clinical accuracy metrics. By synergistically combining explicit difference learning with diagnostic-prior, DoKE provided a powerful and pragmatic approach to automated follow-up CXR interpretation.

Medical foundation models have the potential to revolutionize healthcare by providing robust and generalized representations of medical data. Medical vision-language pre-training has emerged as a promising approach for learning domain-general representations of medical image and text. Current algorithms that exploit global and local alignment between medical image and text could however be marred by redundant information in medical data. To address this issue, we propose a grounded knowledge-enhanced medical vision-language pre-training (GK-MVLP) framework for chest X-ray. In this framework, medical knowledge was grounded to the appropriate anatomical regions by using a transformer-based grounded knowledge-enhanced module for fine-grained alignment between textural features of medical knowledge and the corresponding anatomical region-level visual features. The performance of GK-MVLP was competitive with or exceeded the state of the art on downstream image understanding tasks (chest X-ray disease classification, disease localization), generative task (report generation), and vision-language understanding task (medical visual question-answering). Our results demonstrate the advantage of incorporating grounding mechanism to remove biases and improve the alignment between chest X-ray image and radiology report.

Subcortical stroke induces widespread connectivity changes between cortical and subcortical regions, which may underpin the ensuing behavioral dysexecutive symptoms. This study therefore investigated the cortical structural connectivity that were related to behavioral dysexecutive symptoms using fiber connectivity density (FiCD) mapping, an approach which combines white matter (WM) fiber tractography and cortex reconstruction. The relationships between cortical structural connectivity of significant clusters and its corresponding cortical thickness (CT), and clinical variables were also evaluated based on region-of-interest analysis. Sixty-four subcortical stroke patients with high-resolution T1-weighted imaging and diffusion tensor imaging were enrolled and the behavioral dysexecutive symptoms were assessed using the dysexecutive questionnaire. The FiCD of the left superior parietal gyrus was positively associated with cognitive executive processing (CTT1 time, r = 0.570, p = 0.047; CVFT total correct, r = 0.582, p = 0.047; CVFT total response, r = 0.605, p = 0.040). Similary, the FiCD of the right superior parietal gyrus was also positively associated with cognitive executive processing, (CTT1 time, r = 0.639, p = 0.034). Conversely, negative correlations were observed between the FiCD and CT of the right (r = -0.612 p = 0.045) superior parietal gyrus.

IntroductionMetacognition (MC) impairment is prevalent among stroke survivors but is frequently undiagnosed and untreated. MC impairment hinders stroke survivors’ ability to recognize their deficits, causing them to engage in activities that exceed their capabilities, set unrealistic performance goals and fail to use adaptive compensatory strategies. The present study will evaluate the clinical, neuropsychological and MRI correlates of MC impairment in a cohort of stroke survivors. The secondary objective is to describe the 12-month course of MC impairment.Methods and analysisThe current study is a prospective cohort study. We will recruit 246 subjects. The project duration is 36 months. Subjects and carers will receive a detailed assessment at a research clinic at three, nine and 15 months after stroke onset (T1/T2/T3). The Chinese version of the Self-Awareness of Deficits Interview (SADI) will be completed by each subject. MC impairment is defined as any SADI subscale score of 2 or more. Potential covariate will be measured as well. Tests of executive functioning will be administered as well. Patients will be examined by MRI within 1 week after the onset of stroke. A stepwise logistic regression will be performed to assess the importance of lesions in the regions of interest. To examine neuropsychological functions in MC impairment, regression analysis of the SADI total and subscale scores will be performed using the significantly correlated neuropsychological functions as predictors. To examine the predictors of MC impairment remission, the demographic, clinical and MRI variables of remitters and non-remitters at T2/T3 will be examined by logistic regression.DiscussionThis project will be the first longitudinal study on MC impairment in stroke survivors. The results will shed light on the association between prefrontal cortex and subcortical lesions and MC impairment risk, symptom severity and outcome.

The apolipoprotein E (APOE) ɛ4 allele is the primary genetic risk factor that influences lipid metabolism and contributes to distinctive Alzheimer's disease pathologies, including increased hippocampal atrophy and accelerated cognitive decline. Synaptic dysfunction can occur in APOE4 carriers even before the appearance of any clinical symptoms. Recent evidence has suggested that this genetic risk factor impacts males and females differently. The sex-specific vulnerability for females to cognitive decline, particularly memory, intensifies post-menopause and emphasizes the need for further investigation. White matter abnormalities, APOE4 allele and disruptions in default mode network connectivity serve as early indicators that are crucial for better understanding Alzheimer's disease progression. This study aims to explore relationships between biological sex, APOE4, default mode network-white matter activity and memory function as measured by the Selective Reminding Test. Participants were categorized by risk level on their APOE4 status. Using longitudinal data from the Harvard Aging Brain Study, we examined sex differences in default mode network-white matter engagement among older individuals with and without the APOE4 allele. Our findings demonstrated a significant reduction in default mode network-white matter activity in the right posterior corona radiata in the high-risk group compared to the low-risk group. High-risk females showed reduction in default mode network-white matter activity in the right superior longitudinal fasciculus, which positively correlated with free recall performance, compared to their low-risk counterparts. Unlike females, males showed no significant changes between the low- and high-risk groups. These results underscore the effectiveness of white matter engagement mapping in differentiating longitudinal changes in memory function related to the genetic risk factor APOE4 and biological sex.

Deep learning methods have shown promise in accelerated MRI reconstruction but face significant challenges under domain shifts between training and testing datasets, such as changes in image contrasts, anatomical regions, and acquisition strategies. To address these challenges, we present the first domain generalization framework specifically designed for accelerated MRI reconstruction to robustness across unseen domains. The framework employs progressive strategies to enforce domain invariance, starting with image-level fidelity consistency to ensure robust reconstruction quality across domains, and feature alignment to capture domain-invariant representations. Advancing beyond these foundations, we propose a novel approach enforcing mechanism-level invariance, termed GenCA-MRI, which aligns intrinsic causal relationships within MRI data. We further develop a computational strategy that significantly reduces the complexity of causal alignment, ensuring its feasibility for real-world applications. Extensive experiments validate the framework’s effectiveness, demonstrating both numerical and visual improvements over the baseline algorithm. GenCA-MRI presents the overall best performance, achieving a PSNR improvement up to 2.15 dB on fastMRI and 1.24 dB on IXI dataset at 8× acceleration, with superior performance in preserving anatomical details and mitigating domain-shift problem.

A chest X-ray radiology report describes abnormal findings not only from X-ray obtained at a given examination, but also findings on disease progression or change in device placement with reference to the X-ray from previous examination. Majority of the efforts on automatic generation of radiology report pertain to reporting the former, but not the latter, type of findings. To the best of the authors' knowledge, there is only one work dedicated to generating summary of the latter findings, i.e., follow-up radiology report summary. In this study, we propose a transformer-based framework to tackle this task. Motivated by our observations on the significance of medical lexicon on the fidelity of report summary generation, we introduce two mechanisms to bestow clinical insight to our model, namely disease probability soft guidance and masked entity modeling loss. The former mechanism employs a pretrained abnormality classifier to guide the presence level of specific abnormalities, while the latter directs the model's attention toward medical lexicon. Extensive experiments were conducted to demonstrate that the performance of our model exceeded the state-of-the-art.

BackgroundPoststroke irritability (PSI) is common among stroke survivors and can lead to a poor quality of life, difficulties in social interactions, criticism from caregivers, and caregiver stress. The planned study will evaluate the clinical, neuropsychological, and magnetic resonance imaging (MRI) correlates of PSI in a cohort of stroke survivors. In addition, the study will examine the 15-month progression of PSI.MethodsThis will be a prospective cohort study that will recruit 285 participants. Participants and their caregivers will undergo detailed assessments at a research clinic at 3, 9, and 15 months after stroke onset (T1/T2/T3). The irritability/lability subscale of the Chinese version of the Neuropsychiatric Inventory (CNPI) will be completed by caregivers. Potential covariates will also be measured. Patients will undergo MRI, including diffusion-weighted imaging, within 1 week of stroke onset. A stepwise logistic regression will be performed to evaluate the importance of lesions in the regions of interest (ROIs) along with other significant variables identified in univariate analyses. These analyses will be repeated for patients with and without PSI at T2 and T3. Repeated measures analysis of covariance (ANCOVA) will be used to assess changes in CNPI scores for the entire sample. In ANCOVA analyses, the frequency of infarcts in the ROIs will be treated as the predictor.DiscussionThis will be the first MRI study on PSI in stroke survivors. The findings will provide insights into the association of the orbitofrontal cortex, anterior cingulate cortex, anterior temporal lobe, insula, amygdala, thalamus, and basal ganglia lesions with the risk of PSI.

BackgroundTime-resolved magnetic resonance fingerprinting (MRF), or 4D-MRF, has been demonstrated its feasibility in motion management in radiotherapy (RT). However, the prohibitive long acquisition time is one of challenges of the clinical implementation of 4D-MRF. The shortening of acquisition time causes data insufficiency in each respiratory phase, leading to poor accuracies and consistencies of the predicted tissues' properties of each phase.PurposeTo develop a technique for the reconstruction of multi-phase parametric maps in four-dimensional magnetic resonance fingerprinting (4D-MRF) through the optimization of local T1 and T2 sensitivities.MethodsThe proposed technique employed an iterative optimization to tailor the data arrangement of each phase by manipulation of inter-phase frames, such that the T1 and T2 sensitivities, which were quantified by the modified Minkowski distance, of the truncated signal evolution curve was maximized. The multi-phase signal evolution curves were modified by sliding window reconstruction and inter-phase frame sharing (SWIFS). Motion correction (MC) and dot product matching were sequentially performed on the modified signal evolution and dictionary to reconstruct the multi-parametric maps. The proposed technique was evaluated by numerical simulations using the extended cardiac-torso (XCAT) phantom with regular and irregular breathing patterns, and by in vivo MRF data of three health volunteers and six liver cancer patients acquired at a 3.0 T scanner.ResultsIn simulation study, the proposed SWIFS approach achieved the overall mean absolute percentage error (MAPE) of 8.62% +/- 1.59% and 16.2% +/- 3.88% for the eight-phases T1 and T2 maps, respectively, in the sagittal view with irregular breathing patterns. In contrast, the overall MAPE of T1 and T2 maps generated by the conventional approach with multiple MRF repetitions were 22.1% +/- 11.0% and 30.8% +/- 14.9%, respectively. For in-vivo study, the predicted mean T1 and T2 of liver by the proposed SWIFS approach were 795 ms +/- 38.9 ms and 58.3 ms +/- 11.7 ms, respectively.ConclusionsBoth simulation and in vivo results showed that the approach empowered by T1 and T2 sensitivities optimization and sliding window under the shortened acquisition of MRF had superior performance in the estimation of multi-phase T1 and T2 maps as compared to the conventional approach with oversampling of MRF data.

A chest X-ray radiology report describes abnormal findings not only from X-ray obtained at current examination, but also findings on disease progression or change in device placement with reference to the X-ray from previous examination. Majority of the efforts on automatic generation of radiology report pertain to reporting the former, but not the latter, type of findings. To the best of the authors' knowledge, there is only one work dedicated to generating summary of the latter findings, i.e., follow-up summary. In this study, we therefore propose a transformer-based framework to tackle this task. Motivated by our observations on the significance of medical lexicon on the fidelity of summary generation, we introduce two mechanisms to bestow expert insight to our model, namely expert soft guidance and masked entity modeling loss. The former mechanism employs a pretrained expert disease classifier to guide the presence level of specific abnormalities, while the latter directs the model's attention toward medical lexicon. Extensive experiments were conducted to demonstrate that the performance of our model is competitive with or exceeds the state-of-the-art.

BackgroundLoss of empathy (LoE) is common among stroke survivors, yet often undiagnosed and thus untreated. LoE is related to the loss of a caring marital relationship, higher care burden and poorer quality of life in carers. The present study will evaluate the clinical and MRI correlates of LoE in a cohort of stroke survivors. The secondary objective is to describe the 12-month course of LoE.MethodsThe current study is a prospective cohort study. We will recruit 246 subjects. Subjects and carers will receive a detailed assessment at a research clinic at 3, 9, and 15 months after stroke onset (T1/T2/T3). The Chinese version of the Interpersonal Reactivity Index (IRI), a 28-item personality assessment tool, will be completed by a carer for each subject. LoE is defined as an IRI total score of 39 or less. Patients will be examined by MRI including diffusion weighted imaging (DWI) within 1 week after the onset of stroke. A stepwise logistic regression will be performed to assess the importance of lesions in the regions of interest. To examine the predictors of LoE remission, the demographic, clinical and MRI variables of remitters and non-remitters at T2/T3 will be examined by logistic regression.DiscussionThis project will be the first longitudinal study on LoE in stroke survivors. The results will shed light on the association between prefrontal cortex and subcortical lesions and LoE risk, symptom severity and outcome. The findings will provide data to advance our understanding of the pathogenesis and clinical course of LoE in stroke as well as other neurological conditions. They are thus likely to be applicable to the large population of neurological patients at risk of LoE and should also stimulate further research in this field.

BackgroundPost-stroke behavioral disinhibition (PSBD) is common in stroke survivors and often presents as impulsive, tactless or vulgar behavior. However, it often remains undiagnosed and thus untreated, even though it can lead to a longer length of stay in a rehabilitation facility. The proposed study will aim to evaluate the clinical, neuropsychological and magnetic resonance imaging (MRI) correlates of PSBD in a cohort of stroke survivors and describe its 12-month course.MethodsThis prospective cohort study will recruit 237 patients and will be conducted at the Neurology Unit of the Prince of Wales Hospital. The project duration will be 24 months. The patients will be examined by multiple MRI methods, including diffusion-weighted imaging, within 1 week after stroke onset. The patients and their caregivers will receive a detailed assessment at a research clinic at 3, 9 and 15 months after stroke onset (T1, T2 and T3, respectively). The disinhibition subscale of the Frontal Systems Behavior Scale (FrSBe) will be completed by each subject and caregiver, and scores ≥65 will be considered to indicate PSBD.A stepwise logistic regression will be performed to assess the importance of lesions in the regions of interest (ROIs), together with other significant variables identified in the univariate analyses. For patients with PSBD at T1, the FrSBe disinhibition scores will be compared between the groups of patients with and without ROI infarcts, using covariance analysis. The demographic, clinical and MRI variables of remitters and non-remitters will be examined again at T2 and T3 by logistic regression.DiscussionThis project will be the first MRI study on PSBD in stroke survivors. The results will shed light on the associations of lesions in the orbitofrontal cortex, anterior temporal lobe and subcortical brain structures with the risk of PSBD. The obtained data will advance our understanding of the pathogenesis and clinical course of PSBD in stroke, as well as other neurological conditions. The findings are thus likely to be applicable to the large population of patients with neurological disorders at risk of PSBD and are expected to stimulate further research in this field.

Introduction Structural and functional abnormalities in the cortical–striatal network (CSN) are hypothesised to play a key role in the pathogenesis of neurological disease-associated fatigue. Some small-scale functional MRI (fMRI) studies have suggested that poststroke fatigue (PSF) is related to focal functional connectivity (FC) changes. To date, there has been no published large-scale fMRI study on PSF. This planned study will examine the role of the CSN FC on PSF.Methods and analysis The planned study will be a prospective cohort study conducted at the Neurology Unit of the Prince of Wales Hospital. We will recruit 738 participants. The project duration will be 36 months. A psychiatrist will administer the Fatigue Severity Scale (FSS) at 3 months (P1) following the index stroke. PSF is defined as an FSS Score≥4.0. PSF severity will be defined by the FSS total score at P1. Participants with PSF at P1 will undergo two follow-up assessments at 9 (P2) and 15 (P3) months post stroke. PSF remission at P2 or P3 will be defined as a 50% reduction in FSS. Participants will undergo MRI examinations within 2 weeks of the 3-month poststroke assessment. Structural MRI, resting-state fMRI and diffusion tensor imaging will be performed. FC, structural connectivity, infarcts, cerebral microbleeds and white matter hyperintensities will be analysed. For the primary analysis, the effect of PSF on the FC, structural connectivity and diffusion metrics of CSN of stroke survivors, voxel-wise two-sample t-tests will be performed with FDR correction for multiple comparison and significance level set at p<0.05.Ethics and dissemination Ethical approval was obtained from the Joint Chinese University of Hong Kong–New Territories East Cluster clinical research ethics committee. The study findings will be shared through peer-reviewed journal publications, national and international conferences and social media platforms.

Aging primarily affects memory and executive functions, a relationship that may be underpinned by the fact that almost all adults over 60 years old develop small vessel disease (SVD). The fact that a wide range of neuropathologies could only explain up to 43% of the variation in age-related cognitive impairment suggests that other factors, such as cognitive reserve, may play a role in the brain's resilience against aging-related cognitive decline. This study aims to examine the relationship between structural-functional-connectivity coupling (SFC), and aging, cognitive abilities and reserve, and SVD-related neuropathologies using a cohort of n = 176 healthy elders from the Harvard Aging Brain Study. The SFC is a recently proposed biomarker that reflects the extent to which anatomical brain connections can predict coordinated neural activity. After controlling for the effect of age, sex, and years of education, global SFC, as well as the intra-network SFC of the dorsolateral somatomotor and dorsal attention networks, and the inter-network SFC between dorsolateral somatomotor and frontoparietal networks decreased with age. The global SFC decreased with total cognitive score. There were significant interaction effects between years of education versus white matter hyperintensities and between years of education versus cerebral microbleeds on inter-network SFC. Enlarged perivascular space in basal ganglia was associated with higher inter-network SFC. Our results suggest that cognitive ability is associated with brain coupling at the global level and cognitive reserve with brain coupling at the inter-functional-brain-cluster level with interaction effect from white matter hyperintensities and cerebral microbleed in a cohort of healthy elderlies.

Background and objective: Prediction of poststroke recovery can be expressed by prognostic biomarkers that are related to the pathophysiology of stroke at the cellular and molecular level as well as to the brain structural and functional reserve after stroke at the systems neuroscience level. This study aimed to review potential biomarkers that can predict poststroke functional recovery. Methods: A narrative review was conducted to qualitatively summarize the current evidence on biomarkers used to predict poststroke functional recovery. Results: Neurophysiological measurements and neuroimaging of the brain and a wide diversity of molecules had been used as prognostic biomarkers to predict stroke recovery. Neurophysiological studies using resting-state electroencephalography (EEG) revealed an interhemispheric asymmetry, driven by an increase in low-frequency oscillation and a decrease in high-frequency oscillation in the ipsilesional hemisphere relative to the contralesional side, which was indicative of individual recovery potential. The magnitude of somatosensory evoked potentials and event-related desynchronization elicited by movement in task-related EEG was positively associated with the quantity of recovery. Besides, transcranial magnetic stimulation (TMS) studies revealed the potential values of using motor-evoked potentials (MEP) and TMS-evoked EEG potentials from the ipsilesional motor cortex as prognostic biomarkers. Brain structures measured using magnetic resonance imaging (MRI) have been implicated in stroke outcome prediction. Specifically, the damage to the corticospinal tract (CST) and anatomical motor connections disrupted by stroke lesion predicted motor recovery. In addition, a wide variety of molecular, genetic, and epigenetic biomarkers, including hemostasis, inflammation, tissue remodeling, apoptosis, oxidative stress, infection, metabolism, brain-derived, neuroendocrine, and cardiac biomarkers, etc., were associated with poor functional outcomes after stroke. However, challenges such as mixed evidence and analytical concerns such as specificity and sensitivity have to be addressed before including molecular biomarkers in routine clinical practice. Conclusion: Potential biomarkers with prognostic values for the prediction of functional recovery after stroke have been identified; however, a multimodal approach of biomarkers for prognostic prediction has rarely been studied in the literature. Future studies may incorporate a combination of multiple biomarkers from big data and develop algorithms using data mining methods to predict the recovery potential of patients after stroke in a more precise way.

Purpose: The objective of this study was to develop a respiratory-correlated (RC) 4-dimensional (4D) imaging technique based on magnetic resonance fingerprinting (MRF) (RC-4DMRF) for liver tumor motion management in radiation therapy. Methods and Materials: Thirteen patients with liver cancer were prospectively enrolled in this study. k-space MRF signals of the liver were acquired during free-breathing using the fast acquisition with steady-state precession sequence on a 3T scanner. The signals were binned into 8 respiratory phases based on respiratory surrogates, and interphase displacement vector fields were estimated using a phase-specific low-rank optimization method. Hereafter, the tissue property maps, including T1 and T2 relaxation times, and proton density, were reconstructed using a pyramid motion-compensated method that alternatively optimized interphase displacement vector fields and subspace images. To evaluate the efficacy of RC-4DMRF, amplitude motion differences and Pearson correlation coefficients were determined to assess measurement agreement in tumor motion between RC-4DMRF and cine magnetic resonance imaging (MRI); mean absolute percentage errors of the RC-4DMRF -derived tissue maps were calculated to reveal tissue quantification accuracy using digital human phantom; and tumor-to-liver contrast-to-noise ratio of RC-4DMRF images was compared with that of planning CT and contrast-enhanced MRI (CE-MRI) images. A paired Student t test was used for statistical significance analysis with a P value threshold of .05. Results: RC-4DMRF achieved excellent agreement in motion measurement with cine MRI, yielding the mean (& PLUSMN; standard deviation) Pearson correlation coefficients of 0.95 & PLUSMN; 0.05 and 0.93 & PLUSMN; 0.09 and amplitude motion differences of 1.48 & PLUSMN; 1.06 mm and 0.81 & PLUSMN; 0.64 mm in the superior-inferior and anterior-posterior directions, respectively. Moreover, RC-4DMRF achieved high accuracy in tissue property quantification, with mean absolute percentage errors of 8.8%, 9.6%, and 5.0% for T1, T2, and proton density, respectively. Notably, the tumor contrast-to-noise ratio in RC-4DMRI-derived T1 maps (6.41 & PLUSMN; 3.37) was found to be the highest among all tissue property maps, approximately equal to that of CE-MRI (6.96 & PLUSMN; 1.01, P = .862), and substantially higher than that of planning CT (2.91 & PLUSMN; 1.97, P = .048). Conclusions: RC-4DMRF demonstrated high accuracy in respiratory motion measurement and tissue properties quantifica-tion, potentially facilitating tumor motion management in liver radiation therapy.& COPY; 2023 Elsevier Inc. All rights reserved.

Purpose: To enable 2D Nyquist ghost correction for multi-band interleaved echo-planar imaging (EPI) without reference scan. Methods: 2D phase errors between positive and negative echo were directly measured from the multi-band interleaved EPI data acquired with alternating readout polarity, and then corrected by using multiplexed sensitivity encoding (MUSE) framework. The proposed method was tested on brain data acquired with and without multi-band imaging under different multi-shot factors (4, 8, and 16). In addition, the 2D phase corrections for the interleaved EPI data acquired with 2D zoomed-FOV, 2D reduced-FOV, or 3D multi-band imaging were also performed with the proposed method. The performance of our proposed method was assessed with single-to-noise ratio (SNR) and ghost-to-signal ratio (GSR), and then compared with an iterative 2D phase correction method. The g-factor penalty of the proposed 2D Nyquist ghost correction method associated with 2D phase error was also evaluated. The feasibility of the proposed method was evaluated in diffusion MRI and multi-echo fMRI where quantitative biomarkers with high quality are desired. Results: The proposed method successfully suppressed the ghost artifacts for the multi-shot interleaved EPI data acquired with different multi-shot factors, 2D and 3D multi-band acquisitions, zoomed-FOV, or reduced-FOV. Compared with the iterative method, our proposed method showed lower GSR and comparable SNR performance. Through g-factor penalty simulation, the proposed method showed less than 9% of overall SNR loss associated with a maximum y-linear phase error of π . The residual aliasing artifact was effectively eliminated in the quantitative biomarkers (ADC, FA and dynamic T2*) when applying the proposed method in diffusion imaging and multi-echo fMRI. Conclusion: The proposed method can perform robust 2D Nyquist ghost correction for multi-band multi-shot interleaved EPI without reference scan, and thus can be a generalized 2D ghost correction method for multi-shot interleaved EPI based acquisitions and applications.

Background Cerebral microbleeds (CMB) are indicators of severe cerebral small vessel disease (CSVD) that can be identified through hemosiderin‐sensitive sequences in MRI. Specifically, quantitative susceptibility mapping (QSM) and deep learning were applied to detect CMBs in MRI. Purpose To automatically detect CMB on QSM, we proposed a two‐stage deep learning pipeline. Study Type Retrospective. Subjects A total number of 1843 CMBs from 393 patients (69 ± 12) with cerebral small vessel disease were included in this study. Seventy‐eight subjects (70 ± 13) were used as external testing. Field Strength/Sequence 3 T/QSM. Assessment The proposed pipeline consisted of two stages. In stage I, 2.5D fast radial symmetry transform (FRST) algorithm along with a one‐layer convolutional network was used to identify CMB candidate regions in QSM images. In stage II, the V‐Net was utilized to reduce false positives. The V‐Net was trained using CMB and non CMB labels, which allowed for high‐level feature extraction and differentiation between CMBs and CMB mimics like vessels. The location of CMB was assessed according to the microbleeds anatomical rating scale (MARS) system. Statistical Tests The sensitivity and positive predicative value (PPV) were reported to evaluate the performance of the model. The number of false positive per subject was presented. Results Our pipeline demonstrated high sensitivities of up to 94.9% at stage I and 93.5% at stage II. The overall sensitivity was 88.9%, and the false positive rate per subject was 2.87. With respect to MARS, sensitivities of above 85% were observed for nine different brain regions. Data Conclusion We have presented a deep learning pipeline for detecting CMB in the CSVD cohort, along with a semi‐automated MARS scoring system using the proposed method. Our results demonstrated the successful application of deep learning for CMB detection on QSM and outperformed previous handcrafted methods. Level of Evidence 2 Technical Efficacy Stage 2