Machine learning trained with quantitative susceptibility mapping to detect mild cognitive impairment in Parkinson's disease

MRI-informed machine learning-driven brain age models for classifying mild cognitive impairment converters

BACKGROUND

Brain age model, including estimated brain age and brain-predicted age difference (brain-PAD), has shown great potentials for serving as imaging markers for monitoring normal ageing, as well as for identifying the individuals in the pre-diagnostic phase of neurodegenerative diseases.

PURPOSE

This study aimed to investigate the brain age models in normal ageing and mild cognitive impairments (MCI) converters and their values in classifying MCI conversion.

METHODS

Pre-trained brain age model was constructed using the structural magnetic resonance imaging (MRI) data from the Cambridge Centre for Ageing and Neuroscience (Cam-CAN) project (N = 609). The tested brain age model was built using the baseline, 1-year and 3-year follow-up MRI data from normal ageing (NA) adults (n = 32) and MCI converters (n = 22) drew from the Open Access Series of Imaging Studies (OASIS-2). The quantitative measures of morphometry included total intracranial volume (TIV), gray matter volume (GMV) and cortical thickness. Brain age models were calculated based on the individual's morphometric features using the support vector machine (SVM) algorithm.

RESULTS

With comparable chronological age, MCI converters showed significant increased TIV-based (Baseline: P = 0.021; 1-year follow-up: P = 0.037; 3-year follow-up: P = 0.001) and left GMV-based brain age than NA adults at all time points. Higher brain-PAD scores were associated with worse global cognition. Acceptable classification performance of TIV-based (AUC = 0.698) and left GMV-based brain age (AUC = 0.703) was found, which could differentiate the MCI converters from NA adults at the baseline.

CONCLUSIONS

This is the first demonstration that MRI-informed brain age models exhibit feature-specific patterns. The greater GMV-based brain age observed in MCI converters may provide new evidence for identifying the individuals at the early stage of neurodegeneration. Our findings added value to existing quantitative imaging markers and might help to improve disease monitoring and accelerate personalized treatments in clinical practice.

Are the criteria for PD-MCI diagnosis comprehensive? A Machine Learning study with modified criteria

BACKGROUND

Mild cognitive impairment in Parkinson's disease (PD-MCI) includes deficits in different cognitive domains, and one domain to explore for neurocognitive impairment following the DSM-V is social cognition. However, this domain is not included in current criteria for PD-MCI diagnosis. Moreover, tests vary across studies. It is, therefore, crucial to optimize cognitive assessment in PD-MCI. We aimed to do so by using Machine Learning.

METHODS

275 PD patients were included. Four cognitive batteries were created: two Standard ones (Levels I and II), applying current criteria and "traditional" tests; two Alternative ones (Levels I and II), which incorporated a test of social cognition. These batteries were included in the Random Forest (RF) classifier. To assess RF performance, the AUC was considered, and the Variable Importance Index was estimated to understand the contribution of each test in PD-MCI classification.

RESULTS

Standard Level I and II showed an AUC of 0.852 and 0.892, while Alternative Level I and II showed an AUC of 0.898 and of 0.906. Variable Importance Index revealed that TMT B-A, Ekman test, RAVLT-IR, MoCA, and Action Naming were tests that most contributed to PD-MCI classification.

CONCLUSION

The Alternative level I assessment demonstrated a similar classification capacity to the Standard level II assessment. This finding suggests that in the cognitive assessment of PD patients, it is crucial to consider the most affected cognitive domains in this clinical population, including social cognition. Taken together, these results suggest to revise current criteria for the diagnosis of PD-MCI.

Machine learning for the detection and diagnosis of cognitive impairment in Parkinson's Disease: A systematic review

BACKGROUND

Parkinson's Disease is the second most common neurological disease in over 60s. Cognitive impairment is a major clinical symptom, with risk of severe dysfunction up to 20 years post-diagnosis. Processes for detection and diagnosis of cognitive impairments are not sufficient to predict decline at an early stage for significant impact. Ageing populations, neurologist shortages and subjective interpretations reduce the effectiveness of decisions and diagnoses. Researchers are now utilising machine learning for detection and diagnosis of cognitive impairment based on symptom presentation and clinical investigation. This work aims to provide an overview of published studies applying machine learning to detecting and diagnosing cognitive impairment, evaluate the feasibility of implemented methods, their impacts, and provide suitable recommendations for methods, modalities and outcomes.

METHODS

To provide an overview of the machine learning techniques, data sources and modalities used for detection and diagnosis of cognitive impairment in Parkinson's Disease, we conducted a review of studies published on the PubMed, IEEE Xplore, Scopus and ScienceDirect databases. 70 studies were included in this review, with the most relevant information extracted from each. From each study, strategy, modalities, sources, methods and outcomes were extracted.

RESULTS

Literatures demonstrate that machine learning techniques have potential to provide considerable insight into investigation of cognitive impairment in Parkinson's Disease. Our review demonstrates the versatility of machine learning in analysing a wide range of different modalities for the detection and diagnosis of cognitive impairment in Parkinson's Disease, including imaging, EEG, speech and more, yielding notable diagnostic accuracy.

CONCLUSIONS

Machine learning based interventions have the potential to glean meaningful insight from data, and may offer non-invasive means of enhancing cognitive impairment assessment, providing clear and formidable potential for implementation of machine learning into clinical practice.

Predicting the occurrence of mild cognitive impairment in Parkinson's disease using structural MRI data

Introduction

Mild cognitive impairment (MCI) is a common symptom observed in individuals with Parkinson's disease (PD) and a main risk factor for progressing to dementia. Our objective was to identify early anatomical brain changes that precede the transition from healthy cognition to MCI in PD.

Methods

Structural T1-weighted magnetic resonance imaging data of PD patients with healthy cognition at baseline were downloaded from the Parkinson's Progression Markers Initiative database. Patients were divided into two groups based on the annual cognitive assessments over a 5-year time span: (i) PD patients with unstable healthy cognition who developed MCI over a 5-year follow-up (PD-UHC, n = 52), and (ii) PD patients who maintained stable healthy cognitive function over the same period (PD-SHC, n = 52). These 52 PD-SHC were selected among 192 PD-SHC patients using propensity score matching method to have similar demographic and clinical characteristics with PD-UHC at baseline. Seventy-five percent of these were used to train a support vector machine (SVM) algorithm to distinguish between the PD-UHC and PD-SHC groups, and tested on the remaining 25% of individuals. Shapley Additive Explanations (SHAP) feature analysis was utilized to identify the most informative brain regions in SVM classifier.

Results

The average accuracy of classifying PD-UHC vs. PD-SHC was 80.76%, with 82.05% sensitivity and 79.48% specificity using 10-fold cross-validation. The performance was similar in the hold-out test sets with all accuracy, sensitivity, and specificity at 76.92%. SHAP analysis showed that the most influential brain regions in the prediction model were located in the frontal, occipital, and cerebellar regions as well as midbrain.

Discussion

Our machine learning-based analysis yielded promising results in identifying PD individuals who are at risk of cognitive decline from the earliest disease stage and revealed the brain regions which may be linked to the prospective cognitive decline in PD before clinical symptoms emerge.

Parkinson's disease and Parkinsonism syndromes: Evaluating iron deposition in the putamen using magnetic susceptibility MRI techniques - A systematic review and literature analysis

Magnetic resonance imaging (MRI) techniques, such as quantitative susceptibility mapping (QSM) and susceptibility-weighted imaging (SWI), can detect iron deposition in the brain. Iron accumulation in the putamen (PUT) can contribute to the pathogenesis of Parkinson's disease (PD) and atypical Parkinsonian disorders. This systematic review aimed to synthesize evidence on iron deposition in the PUT assessed by MRI susceptibility techniques in PD and Parkinsonism syndromes. The PubMed and Scopus databases were searched for relevant studies. Thirty-four studies from January 2007 to October 2023 that used QSM, SWI, or other MRI susceptibility methods to measure putaminal iron in PD, progressive supranuclear palsy (PSP), multiple system atrophy (MSA), and healthy controls (HCs) were included. Most studies have found increased putaminal iron levels in PD patients versus HCs based on higher quantitative susceptibility. Putaminal iron accumulation correlates with worse motor scores and cognitive decline in patients with PD. Evidence regarding differences in susceptibility between PD and atypical Parkinsonism is emerging, with several studies showing greater putaminal iron deposition in PSP and MSA than in PD patients. Alterations in putaminal iron levels help to distinguish these disorders from PD. Increased putaminal iron levels appear to be associated with increased disease severity and progression. Thus, magnetic susceptibility MRI techniques can detect abnormal iron accumulation in the PUT of patients with Parkinsonism. Moreover, quantifying putaminal susceptibility may serve as an MRI biomarker to monitor motor and cognitive changes in PD and aid in the differential diagnosis of Parkinsonian disorders.

Neuroimaging of Parkinson's disease by quantitative susceptibility mapping

Parkinson's disease (PD) is a common neurodegenerative disease, and apart from a few rare genetic causes, its pathogenesis remains largely unclear. Recent scientific interest has been captured by the involvement of iron biochemistry and the disruption of iron homeostasis, particularly within the brain regions specifically affected in PD. The advent of Quantitative Susceptibility Mapping (QSM) has enabled non-invasive quantification of brain iron in vivo by MRI, which has contributed to the understanding of iron-associated pathogenesis and has the potential for the development of iron-based biomarkers in PD. This review elucidates the biochemical underpinnings of brain iron accumulation, details advancements in iron-sensitive MRI technologies, and discusses the role of QSM as a biomarker of iron deposition in PD. Despite considerable progress, several challenges impede its clinical application after a decade of QSM studies. The initiation of multi-site research is warranted for developing robust, interpretable, and disease-specific biomarkers for monitoring PD disease progression.

Increased brain iron in patients with thyroid-associated ophthalmopathy: a whole-brain analysis

Background

To investigate the whole-brain iron deposition alternations in patients with thyroid-associated ophthalmopathy (TAO) using quantitative susceptibility mapping (QSM).

Methods

Forty-eight patients with TAO and 33 healthy controls (HCs) were enrolled. All participants underwent brain magnetic resonance imaging scans and clinical scale assessments. QSM values were calculated and compared between TAO and HCs groups using a voxel-based analysis. A support vector machine (SVM) analysis was performed to evaluate the performance of QSM values in differentiating patients with TAO from HCs.

Results

Compared with HCs, patients with TAO showed significantly increased QSM values in the bilateral caudate nucleus (CN), left thalamus (TH), left cuneus, left precuneus, right insula and right middle frontal gyrus. In TAO group, QSM values in left TH were positively correlated with Hamilton Depression Rating Scale (HDRS) scores (r = 0.414, p = 0.005). The QSM values in right CN were negatively correlated with Montreal Cognitive Assessment (MoCA) scores (r = -0.342, p = 0.021). Besides that, a nearly negative correlation was found between QSM values in left CN and MoCA scores (r = -0.286, p = 0.057). The SVM model showed a good performance in distinguishing patients with TAO from the HCs (area under the curve, 0.958; average accuracy, 90.1%).

Conclusion

Patients with TAO had significantly increased iron deposition in brain regions corresponding to known visual, emotional and cognitive deficits. QSM values could serve as potential neuroimaging markers of TAO.

Pulvinar quantitative susceptibility mapping predicts visual hallucinations post-deep brain stimulation in Parkinson's disease

PURPOSE

We have reported the relationship between low pulvinar nuclei (PN) intensity in susceptibility-weighted imaging and the appearance of visual hallucinations and cognitive function. The aim of the study was to examine the changes in the quantitative susceptibility mapping (QSM) in patients with Parkinson's disease (PD) who underwent deep brain stimulation (DBS) and verify whether the PN susceptibility value (SV) on QSM can predict visual hallucination and cognitive changes after DBS.

METHODS

This study examined 24 patients with PD who underwent DBS along with QSM imaging on magnetic resonance imaging (MRI). All MRIs were performed within 3 months before surgery. The PN SV was further assessed based on the QSM. Then, associations were examined among cognitive changes, hallucination, and PN SV. The cognitive function of the patient was compared immediately before surgery and at 1 year postoperatively.

RESULTS

Visual hallucinations were observed in seven patients during the follow-up period. The PN SV was ≥0.045 ppm in nine patients with PD, and six of them had visual hallucinations, whereas only one of 15 patients with PD with SV of <0.045 ppm had visual hallucinations (Fisher's exact test, p = .0037).

CONCLUSIONS

The SV of >0.045 ppm at the PN in QSM in patients with PD may provide useful information suggesting visual hallucination and cognitive deterioration after DBS treatment.

Assessing white matter microstructural changes in idiopathic normal pressure hydrocephalus using voxel-based R2* relaxometry analysis

Background

R2* relaxometry and quantitative susceptibility mapping can be combined to distinguish between microstructural changes and iron deposition in white matter. Here, we aimed to explore microstructural changes in the white matter associated with clinical presentations such as cognitive impairment in patients with idiopathic normal-pressure hydrocephalus (iNPH) using R2* relaxometry analysis in combination with quantitative susceptibility mapping.

Methods

We evaluated 16 patients clinically diagnosed with possible or probable iNPH and 18 matched healthy controls (HC) who were chosen based on similarity in age and sex. R2* and quantitative susceptibility mapping were compared using voxel-wise and atlas-based one-way analysis of covariance (ANCOVA). Finally, partial correlation analyses were performed to assess the relationship between R2* and clinical presentations.

Results

R2* was lower in some white matter regions, including the bilateral superior longitudinal fascicle and sagittal stratum, in the iNPH group compared to the HC group. The voxel-based quantitative susceptibility mapping results did not differ between the groups. The atlas-based group comparisons yielded negative mean susceptibility values in almost all brain regions, indicating no clear paramagnetic iron deposition in the white matter of any subject. R2* and cognitive performance scores between the left superior longitudinal fasciculus (SLF) and right sagittal stratum (SS) were positively correlated. In addition to that, R2* and gait disturbance scores between left SS were negatively correlated.

Conclusion

Our analysis highlights the microstructural changes without iron deposition in the SLF and SS, and their association with cognitive impairment and gait disturbance in patients with iNPH.

Detection of mild cognitive impairment in Parkinson's disease using gradient boosting decision tree models based on multilevel DTI indices

BACKGROUND

Cognitive dysfunction is the most common non-motor symptom in Parkinson's disease (PD), and timely detection of a slight cognitive decline is crucial for early treatment and prevention of dementia. This study aimed to build a machine learning model based on intra- and/or intervoxel metrics extracted from diffusion tensor imaging (DTI) to automatically classify PD patients without dementia into mild cognitive impairment (PD-MCI) and normal cognition (PD-NC) groups.

METHODS

We enrolled PD patients without dementia (52 PD-NC and 68 PD-MCI subtypes) who were assigned to the training and test datasets in an 8:2 ratio. Four intravoxel metrics, including fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD), and two novel intervoxel metrics, local diffusion homogeneity (LDH) using Spearman's rank correlation coefficient (LDHs) and Kendall's coefficient concordance (LDHk), were extracted from the DTI data. Decision tree, random forest, and eXtreme gradient boosting (XGBoost) models based on individual and combined indices were built for classification, and model performance was assessed and compared via the area under the receiver operating characteristic curve (AUC). Finally, feature importance was evaluated using SHapley Additive exPlanation (SHAP) values.

RESULTS

The XGBoost model based on a combination of the intra- and intervoxel indices achieved the best classification performance, with an accuracy of 91.67%, sensitivity of 92.86%, and AUC of 0.94 in the test dataset. SHAP analysis showed that the LDH of the brainstem and MD of the right cingulum (hippocampus) were important features.

CONCLUSIONS

More comprehensive information on white matter changes can be obtained by combining intra- and intervoxel DTI indices, improving classification accuracy. Furthermore, machine learning methods based on DTI indices can be used as alternatives for the automatic identification of PD-MCI at the individual level.

Long-term cognitive decline prediction based on multi-modal data using Multimodal3DSiameseNet: transfer learning from Alzheimer's disease to Parkinson's disease

PURPOSE

Monitoring and predicting the cognitive state of subjects with neurodegenerative disorders is crucial to provide appropriate treatment as soon as possible. In this work, we present a machine learning approach using multimodal data (brain MRI and clinical) from two early medical visits, to predict the longer-term cognitive decline of patients. Using transfer learning, our model can be successfully transferred from one neurodegenerative disease (Alzheimer's) to another (Parkinson's).

METHODS

Our model is a Deep Neural Network with siamese sub-modules dedicated to extracting features from each modality. We pre-train it with data from ADNI (Alzheimer's disease), then transfer it on the smaller PPMI dataset (Parkinson's disease). We show that, even when we do not fine-tune the filters learnt from the ADNI MRIs, the transferred model's results are satisfying on PPMI.

RESULTS

The first main result is that our model provides satisfying long-term predictions of cognitive decline from any pair of early visits, with no fixed time delay between these visits (provided the potential decline has started at the second visit). The second main result is that the prediction performance on Parkinson's dataset (PPMI) reaches an AUC of 0.81 on PPMI after transfer learning from Alzheimer's dataset (ADNI), without even having to re-train the image filters, versus an AUC of 0.72 for the model trained from scratch on PPMI.

CONCLUSIONS

First, our model is effective for predicting long-term cognitive decline from only two visits, even with irregular intervals of time. When dealing with neurodegenerative diseases, where patients often miss some control visits, this is an important finding. Second, our model is able to transfer the knowledge learnt from one neurodegenerative disease (Alzheimer's) to another (Parkinson's), when using the same imaging modalities (brain MRI) and different clinical variables. This makes it usable even for diseases that are rare or under-studied.

Quantitative susceptibility mapping as an imaging biomarker for Alzheimer’s disease: The expectations and limitations

Alzheimer’s disease (AD) is the most common type of dementia and a distressing diagnosis for individuals and caregivers. Researchers and clinical trials have mainly focused on β-amyloid plaques, which are hypothesized to be one of the most important factors for neurodegeneration in AD. Meanwhile, recent clinicopathological and radiological studies have shown closer associations of tau pathology rather than β-amyloid pathology with the onset and progression of Alzheimer’s symptoms. Toward a biological definition of biomarker-based research framework for AD, the 2018 National Institute on Aging–Alzheimer’s Association working group has updated the ATN classification system for stratifying disease status in accordance with relevant pathological biomarker profiles, such as cerebral β-amyloid deposition, hyperphosphorylated tau, and neurodegeneration. In addition, altered iron metabolism has been considered to interact with abnormal proteins related to AD pathology thorough generating oxidative stress, as some prior histochemical and histopathological studies supported this iron-mediated pathomechanism. Quantitative susceptibility mapping (QSM) has recently become more popular as a non-invasive magnetic resonance technique to quantify local tissue susceptibility with high spatial resolution, which is sensitive to the presence of iron. The association of cerebral susceptibility values with other pathological biomarkers for AD has been investigated using various QSM techniques; however, direct evidence of these associations remains elusive. In this review, we first briefly describe the principles of QSM. Second, we focus on a large variety of QSM applications, ranging from common applications, such as cerebral iron deposition, to more recent applications, such as the assessment of impaired myelination, quantification of venous oxygen saturation, and measurement of blood– brain barrier function in clinical settings for AD. Third, we mention the relationships among QSM, established biomarkers, and cognitive performance in AD. Finally, we discuss the role of QSM as an imaging biomarker as well as the expectations and limitations of clinically useful diagnostic and therapeutic implications for AD.