FDG-PET combined with learning vector quantization allows classification of neurodegenerative diseases and reveals the trajectory of idiopathic REM sleep behavior disorder

IRMA: Machine learning-based harmonization of 18 F-FDG PET brain scans in multi-center studies

PURPOSE

Center-specific effects in PET brain scans arise due to differences in technical and procedural aspects. This restricts the merging of data between centers and introduces source-specific bias.

METHODS

We demonstrate the use of the recently proposed machine learning method Iterated Relevance Matrix Analysis (IRMA) for harmonization of center-specific effects in brain18 F-Fluorodeoxyglucose (18 F-FDG) PET scans. The center difference is learned by applying IRMA on PCA-based feature vectors of healthy controls (HC), resulting in a subspace V , representing information not comparable between centers, and the remaining subspace U , where no center differences are present. In this proof-of-concept study, we demonstrate the properties of the method using data from four centers. After center-harmonization, a Generalized Matrix Learning Vector Quantization (GMLVQ) model was trained to discriminate between Parkinson's disease, Alzheimer's disease and Dementia with Lewy Bodies.

RESULTS

At the initial IRMA iteration, the system was able to determine the center origin of the four HC cohorts almost perfectly. The method required six iterations, corresponding to a six-dimensional subspace V , to determine the entire center difference. An uncorrected disease classification model was highly biased to center-specific effects, creating a falsely inflated performance when applying internal (cross-) validation. The cross-validation performance of the center-harmonized model remained high, while it generalized significantly better to unseen test cohorts. Furthermore, the framework is highly transparent, providing analytic reconstructions of the correction and visualizations of the data in voxel space.

CONCLUSION

IRMA can be used to learn and disregard center-specific information in features extracted from brain18 F-FDG PET scans, while retaining disease-specific information.

Explainable machine learning for movement disorders - Classification of tremor and myoclonus

BACKGROUND

Treatment for essential tremor (ET) and cortical myoclonus (CM) differs. As their clinical distinction can be difficult, with large inter- and intra-observer variability, there is a need for additional diagnostic tools.

OBJECTIVES

We aim to develop a machine learning based approach using accelerometry data to support clinicians in their diagnosis of ET versus CM.

METHODS

We recorded the upper body movements of 19 ET and 19 CM patients performing 21 tasks using eight accelerometry sensors. The phenotype was classified by applying an explainable machine learning method called generalized matrix learning vector quantization (GMLVQ) to the power spectrum derived from the accelerometry recordings. In addition to the general classification performance, GMLVQ provides context for the decisions made. This includes the relevance of frequencies and thus the identification of distinct patterns in the phenotypes.

RESULTS

For all postures and dynamic tasks, excellent classification results were reached with AUROC approaching 1,0. For the classification, GMLVQ considers the overall shape of the power spectrum, whereby the frequencies 5-7 Hz as well as 3-4 and 9-10 Hz are of particular importance. These frequencies correspond to the tremor peak and local minima that are also observed as characteristics in the literature.

CONCLUSION

The excellent classification results provide a proof of concept for the discrimination of ET and CM by applying a GMLVQ machine learning analysis to power spectra derived from accelerometry recordings. Such a system could support clinicians to improve the diagnostic accuracy.

Brain age prediction based on brain region volume modeling under broad network field of view

BACKGROUND AND OBJECTIVE

Brain region volume from Structural Magnetic Resonance Imaging (sMRI) can directly reflect abnormal states in brain aging. While promising for clinical brain health assessment, existing volume-based brain age prediction methods fail to explore both linear and nonlinear relationships, resulting in weak representation and suboptimal estimates.

METHODS

This paper proposes a brain age prediction method, RFBLSO, based on Random Forest (RF), Broad Learning System (BLS), and Leave-One-Out Cross Validation (LOO). Firstly, RF is used to eliminate redundant brain regions with low correlation to the target value. The objective function is constructed by integrating feature nodes, enhancement nodes, and optimal regularization parameters. Subsequently, the pseudo-inverse method is employed to solve for the output coefficients, which facilitates a more accurate representation of the linear and nonlinear relationships between volume features and brain age.

RESULTS

Across various datasets, RFBLSO demonstrates the capability to formulate brain age prediction models, achieving a Mean Absolute Error (MAE) of 4.60 years within the Healthy Group and 4.98 years within the Chinese2020 dataset. In the Clinical Group, RFBLSO achieves measurement and effective differentiation among Healthy Controls (HC), Mild Cognitive Impairment (MCI), and Alzheimer's disease (AD) (MAE for HC, MCI, and AD: 4.46 years, 8.77 years, 13.67 years; the effect size η2 of the analysis of variance for AD/MCI vs. HC is 0.23; the effect sizes of post-hoc tests are Cohen's d = 0.74 (AD vs. MCI), 1.50 (AD vs. HC), 0.77 (MCI vs. HC)). Compared to other linear or nonlinear brain age prediction methods, RFBLSO offers more accurate measurements and effectively distinguishes between Clinical Groups. This is because RFBLSO can simultaneously explore both linear and nonlinear relationships between brain region volume and brain age.

CONCLUSION

The proposed RFBLSO effectively represents both linear and nonlinear relationships between brain region volume and brain age, allowing for more accurate individual brain age estimation. This provides a feasible method for predicting the risk of neurodegenerative diseases.

Sleep Disturbances in Autoimmune Neurological Diseases: Mechanisms, Clinical Characteristics, Assessment, and Treatment Strategies

PURPOSE OF REVIEW

Sleep disturbances are a hallmark feature of various autoimmune neurological diseases (AINDs). However, limited awareness of these sleep manifestations exists among clinicians. We provide a comprehensive overview of assessment methods, characteristic sleep disturbances, the impact of specific antibodies on sleep patterns, and treatment strategies for sleep disturbances in AINDs.

RECENT FINDINGS

Research advancements in sleep disturbances in autoimmune neurological disease focus primarily on four areas: mechanisms, clinical characteristics, assessment, and treatment. Regarding mechanisms, animal models for AINDs, particularly those involving specific antibodies like anti-NMDAR, anti-LGI1, and anti-IgLON5, have become more comprehensive. Recent advancements in animal models have led to the establishment of numerous models for AINDs; these models include a wide range of antibodies, including anti-NMDAR, anti-LGI1, and anti-IgLON5. Several studies using these models have revealed common mechanisms underlying sleep disturbances in these diseases. In terms of clinical characteristics, the identification of antibodies associated with recently discovered AINDs has expanded the spectrum of sleep disturbance symptoms observed compared to prior findings. A comprehensive evaluation system for the assessment of sleep disturbances has been established, including questionnaires, polysomnography, functional magnetic resonance imaging, and 18F-FDG PET/CT. Additionally, cardiopulmonary coupling shows promise as a novel assessment tool. Currently, no universally effective treatment exists for sleep disturbances in autoimmune neurological diseases, either through symptomatic treatment or immunosuppressive therapy. Further studies are needed to confirm the efficacy of new therapies and validate the benefits of existing treatments. Sleep disturbances are a hallmark feature of AINDs. Recent advancements have significantly expanded our understanding of their assessment and treatment. However, further studies are needed to address the remaining uncertainties in sleep disturbance management.

Next move in movement disorders: neuroimaging protocols for hyperkinetic movement disorders

Introduction

The Next Move in Movement Disorders (NEMO) study is an initiative aimed at advancing our understanding and the classification of hyperkinetic movement disorders, including tremor, myoclonus, dystonia, and myoclonus-dystonia. The study has two main objectives: (a) to develop a computer-aided tool for precise and consistent classification of these movement disorder phenotypes, and (b) to deepen our understanding of brain pathophysiology through advanced neuroimaging techniques. This protocol review details the neuroimaging data acquisition and preprocessing procedures employed by the NEMO team to achieve these goals.

Methods and analysis

To meet the study's objectives, NEMO utilizes multiple imaging techniques, including T1-weighted structural MRI, resting-state fMRI, motor task fMRI, and 18F-FDG PET scans. We will outline our efforts over the past 4 years to enhance the quality of our collected data, and address challenges such as head movements during image acquisition, choosing acquisition parameters and constructing data preprocessing pipelines. This study is the first to employ these neuroimaging modalities in a standardized approach contributing to more uniformity in the analyses of future studies comparing these patient groups. The data collected will contribute to the development of a machine learning-based classification tool and improve our understanding of disorder-specific neurobiological factors.

Ethics and dissemination

Ethical approval has been obtained from the relevant local ethics committee. The NEMO study is designed to pioneer the application of machine learning of movement disorders. We expect to publish articles in multiple related fields of research and patients will be informed of important results via patient associations and press releases.

Subspace corrected relevance learning with application in neuroimaging

In machine learning, data often comes from different sources, but combining them can introduce extraneous variation that affects both generalization and interpretability. For example, we investigate the classification of neurodegenerative diseases using FDG-PET data collected from multiple neuroimaging centers. However, data collected at different centers introduces unwanted variation due to differences in scanners, scanning protocols, and processing methods. To address this issue, we propose a two-step approach to limit the influence of center-dependent variation on the classification of healthy controls and early vs. late-stage Parkinson's disease patients. First, we train a Generalized Matrix Learning Vector Quantization (GMLVQ) model on healthy control data to identify a "relevance space" that distinguishes between centers. Second, we use this space to construct a correction matrix that restricts a second GMLVQ system's training on the diagnostic problem. We evaluate the effectiveness of this approach on the real-world multi-center datasets and simulated artificial dataset. Our results demonstrate that the approach produces machine learning systems with reduced bias - being more specific due to eliminating information related to center differences during the training process - and more informative relevance profiles that can be interpreted by medical experts. This method can be adapted to similar problems outside the neuroimaging domain, as long as an appropriate "relevance space" can be identified to construct the correction matrix.

99mTc-HMPAO SPECT Perfusion Signatures Associated With Clinical Progression in Patients With Isolated REM Sleep Behavior Disorder

BACKGROUND AND OBJECTIVES

Idiopathic/isolated REM sleep behavior disorder (iRBD) is associated with dementia with Lewy bodies and Parkinson disease. Despite evidence of abnormal cerebral perfusion in iRBD, there is currently no pattern that can predict whether an individual will develop dementia with Lewy bodies or Parkinson disease. The objective was to identify a perfusion signature associated with conversion to dementia with Lewy bodies in iRBD.

METHODS

Patients with iRBD underwent video-polysomnography, neurologic and neuropsychological assessments, and baseline 99mTc-HMPAO SPECT to assess relative cerebral blood flow. Partial least squares correlation was used to identify latent variables that maximized covariance between 27 clinical features and relative gray matter perfusion. Patient-specific scores on the latent variables were used to test the association with conversion to dementia with Lewy bodies compared with that with Parkinson disease. The signature's expression was also assessed in 24 patients with iRBD who underwent a second perfusion scan, 22 healthy controls, and 19 individuals with Parkinson disease.

RESULTS

Of the 137 participants, 93 underwent SPECT processing, namely 52 patients with iRBD (67.9 years, 73% men), 19 patients with Parkinson disease (67.3 years, 37% men), and 22 controls (67.0 years, 73% men). Of the 47 patients with iRBD followed up longitudinally (4.5 years), 12 (26%) developed a manifest synucleinopathy (4 dementia with Lewy bodies and 8 Parkinson disease). Analysis revealed 2 latent variables between relative blood flow and clinical features: the first was associated with a broad set of features that included motor, cognitive, and perceptual variables, age, and sex; the second was mostly associated with cognitive features and RBD duration. When brought back into the patient's space, the expression of the first variable was associated with conversion to a manifest synucleinopathy, whereas the second was associated with conversion to dementia with Lewy bodies. The expression of the patterns changed over time and was associated with worse motor features.

DISCUSSION

This study identified a brain perfusion signature associated with cognitive impairment in iRBD and transition to dementia with Lewy bodies. This signature, which can be derived from individual scans, has the potential to be developed into a biomarker that predicts dementia with Lewy bodies in at-risk individuals.

Deep learning based diagnosis of Alzheimer's disease using FDG-PET images

PURPOSE

The aim of this study is to develop a deep neural network to diagnosis Alzheimer's disease and categorize the stages of the disease using FDG-PET scans. Fluorodeoxyglucose positron emission tomography (FDG-PET) is a highly effective diagnostic tool that accurately detects glucose metabolism in the brain of AD patients.

MATERIAL AND METHODS

In this work, we have developed a deep neural network using FDG-PET to discriminate Alzheimer's disease subjects from stable mild cognitive impairment (sMCI), progressive mild cognitive impairment (pMCI), and cognitively normal (CN) cohorts. A total of 83 FDG-PET scans are collected from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database, including 21 subjects with CN, 21 subjects with sMCI, 21 subjects with pMCI, and 20 subjects with AD.

RESULTS

The method has achieved remarkable accuracy rates of 99.31% for CN vs. AD, 99.88% for CN vs. MCI, 99.54% for AD vs. MCI, and 96.81% for pMCI vs. sMCI. Based on the experimental results.

CONCLUSION

The results show that the proposed method has a significant generalisation ability as well as good performance in predicting the conversion of MCI to AD even in the absence of direct information. FDG-PET is a well-known biomarker for the identification of Alzheimer's disease using transfer learning.

Comparison of univariate and multivariate analyses for brain [18F]FDG PET data in α-synucleinopathies

BACKGROUND

Brain imaging with [18F]FDG-PET can support the diagnostic work-up of patients with α-synucleinopathies. Validated data analysis approaches are necessary to evaluate disease-specific brain metabolism patterns in neurodegenerative disorders. This study compared the univariate Statistical Parametric Mapping (SPM) single-subject procedure and the multivariate Scaled Subprofile Model/Principal Component Analysis (SSM/PCA) in a cohort of patients with α-synucleinopathies.

METHODS

We included [18F]FDG-PET scans of 122 subjects within the α-synucleinopathy spectrum: Parkinson's Disease (PD) normal cognition on long-term follow-up (PD - low risk to dementia (LDR); n = 28), PD who developed dementia on clinical follow-up (PD - high risk of dementia (HDR); n = 16), Dementia with Lewy Bodies (DLB; n = 67), and Multiple System Atrophy (MSA; n = 11). We also included [18F]FDG-PET scans of isolated REM sleep behaviour disorder (iRBD; n = 51) subjects with a high risk of developing a manifest α-synucleinopathy. Each [18F]FDG-PET scan was compared with 112 healthy controls using SPM procedures. In the SSM/PCA approach, we computed the individual scores of previously identified patterns for PD, DLB, and MSA: PD-related patterns (PDRP), DLBRP, and MSARP. We used ROC curves to compare the diagnostic performances of SPM t-maps (visual rating) and SSM/PCA individual pattern scores in identifying each clinical condition across the spectrum. Specifically, we used the clinical diagnoses ("gold standard") as our reference in ROC curves to evaluate the accuracy of the two methods. Experts in movement disorders and dementia made all the diagnoses according to the current clinical criteria of each disease (PD, DLB and MSA).

RESULTS

The visual rating of SPM t-maps showed higher performance (AUC: 0.995, specificity: 0.989, sensitivity 1.000) than PDRP z-scores (AUC: 0.818, specificity: 0.734, sensitivity 1.000) in differentiating PD-LDR from other α-synucleinopathies (PD-HDR, DLB and MSA). This result was mainly driven by the ability of SPM t-maps to reveal the limited or absent brain hypometabolism characteristics of PD-LDR. Both SPM t-maps visual rating and SSM/PCA z-scores showed high performance in identifying DLB (DLBRP = AUC: 0.909, specificity: 0.873, sensitivity 0.866; SPM t-maps = AUC: 0.892, specificity: 0.872, sensitivity 0.910) and MSA (MSARP: AUC: 0.921, specificity: 0.811, sensitivity 1.000; SPM t-maps: AUC: 1.000, specificity: 1.000, sensitivity 1.000) from other α-synucleinopathies. PD-HDR and DLB were comparable for the brain hypo and hypermetabolism patterns, thus not allowing differentiation by SPM t-maps or SSM/PCA. Of note, we found a gradual increase of PDRP and DLBRP expression in the continuum from iRBD to PD-HDR and DLB, where the DLB patients had the highest scores. SSM/PCA could differentiate iRBD from DLB, reflecting specifically the differences in disease staging and severity (AUC: 0.938, specificity: 0.821, sensitivity 0.941).

CONCLUSIONS

SPM-single subject maps and SSM/PCA are both valid methods in supporting diagnosis within the α-synucleinopathy spectrum, with different strengths and pitfalls. The former reveals dysfunctional brain topographies at the individual level with high accuracy for all the specific subtype patterns, and particularly also the normal maps; the latter provides a reliable quantification, independent from the rater experience, particularly in tracking the disease severity and staging. Thus, our findings suggest that differences in data analysis approaches exist and should be considered in clinical settings. However, combining both methods might offer the best diagnostic performance.

Syndrome Pattern Recognition Method Using Sensed Patient Data for Neurodegenerative Disease Progression Identification

Neurodegenerative diseases are a group of conditions that involve the progressive loss of function of neurons in the brain and spinal cord. These conditions can result in a wide range of symptoms, such as difficulty with movement, speech, and cognition. The causes of neurodegenerative diseases are poorly understood, but many factors are believed to contribute to the development of these conditions. The most important risk factors include ageing, genetics, abnormal medical conditions, toxins, and environmental exposures. A slow decline in visible cognitive functions characterises the progression of these diseases. If left unattended or unnoticed, disease progression can result in serious issues such as the cessation of motor function or even paralysis. Therefore, early recognition of neurodegenerative diseases is becoming increasingly important in modern healthcare. Many sophisticated artificial intelligence technologies are incorporated into modern healthcare systems for the early recognition of these diseases. This research article introduces a Syndrome-dependent Pattern Recognition Method for the early detection and progression monitoring of neurodegenerative diseases. The proposed method determines the variance between normal and abnormal intrinsic neural connectivity data. The observed data is combined with previous and healthy function examination data to identify the variance. In this combined analysis, deep recurrent learning is exploited by tuning the analysis layer based on variance suppressed by identifying normal and abnormal patterns in the combined analysis. This variance from different patterns is recurrently used to train the learning model for maximising of recognition accuracy. The proposed method achieves 16.77% high accuracy, 10.55% high precision, and 7.69% high pattern verification. It reduces the variance and verification time by 12.08% and 12.02%, respectively.