scSTAR reveals hidden heterogeneity with a real-virtual cell pair structure across conditions in single-cell RNA sequencing data.

Cell-state transition can reveal additional information from single-cell ribonucleic acid (RNA)-sequencing data in time-resolved biological phenomena. However, most of the current methods are based on the time derivative of the gene expression state, which restricts them to the short-term evolution of cell states. Here, we present single-cell State Transition Across-samples of RNA-seq data (scSTAR), which overcomes this limitation by constructing a paired-cell projection between biological conditions with an arbitrary time span by maximizing the covariance between two feature spaces using partial least square and minimum squared error methods. In mouse ageing data, the response to stress in CD4+ memory T cell subtypes was found to be associated with ageing. A novel Treg subtype characterized by mTORC activation was identified to be associated with antitumour immune suppression, which was confirmed by immunofluorescence microscopy and survival analysis in 11 cancers from The Cancer Genome Atlas Program. On melanoma data, scSTAR improved immunotherapy-response prediction accuracy from 0.8 to 0.96.

Multi-dimensional relationships among dementia, depression and prescribed drugs in England and Wales hospitals.

BACKGROUND: Dementia is a group of symptoms that largely affects older people. The majority of patients face behavioural and psychological symptoms (BPSD) during the course of their illness. Alzheimer's disease (AD) and vascular dementia (VaD) are two of the most prevalent types of dementia. Available medications provide symptomatic benefits and provide relief from BPSD and associated health issues. However, it is unclear how specific dementia, antidepressant, antipsychotic, antianxiety, and mood stabiliser drugs, used in the treatment of depression and dementia subtypes are prescribed in hospital admission, during hospital stay, and at the time of discharge. To address this, we apply multi-dimensional data analytical approaches to understand drug prescribing practices within hospitals in England and Wales. METHODS: We made use of the UK National Audit of Dementia (NAD) dataset and pre-processed the dataset. We evaluated the pairwise Pearson correlation of the dataset and selected key data features which are highly correlated with dementia subtypes. After that, we selected drug prescribing behaviours (e.g. specific medications at the time of admission, during the hospital stay, and upon discharge), drugs and disorders. Then to shed light on the relations across multiple features or dimensions, we carried out multiple regression analyses, considering the number of dementia, antidepressant, antipsychotic, antianxiety, mood stabiliser, and antiepileptic/anticonvulsant drug prescriptions as dependent variables, and the prescription of other drugs, number of patients with dementia subtypes (AD/VaD), and depression as independent variables. RESULTS: In terms of antidepressant drugs prescribed in hospital admission, during stay and discharge, the number of sertraline and venlafaxine prescriptions were associated with the number of VaD patients whilst the number of mirtazapine prescriptions was associated with frontotemporal dementia patients. During admission, the number of lamotrigine prescriptions was associated with frontotemporal dementia patients, and with the number of valproate and dosulepin prescriptions. During discharge, the number of mirtazapine prescriptions was associated with the number of donepezil prescriptions in conjunction with frontotemporal dementia patients. Finally, the number of prescriptions of donepezil/memantine at admission, during hospital stay and at discharge exhibited positive association with AD patients. CONCLUSION: Our analyses reveal a complex, multifaceted set of interactions among prescribed drug types, dementia subtypes, and depression.

Changes-of-mind in the absence of new post-decision evidence.

Decisions are occasionally accompanied by changes-of-mind. While considered a hallmark of cognitive flexibility, the mechanisms underlying changes-of-mind remain elusive. Previous studies on perceptual decision making have focused on changes-of-mind that are primarily driven by the accumulation of additional noisy sensory evidence after the initial decision. In a motion discrimination task, we demonstrate that changes-of-mind can occur even in the absence of additional evidence after the initial decision. Unlike previous studies of changes-of-mind, the majority of changes-of-mind in our experiment occurred in trials with prolonged initial response times. This suggests a distinct mechanism underlying such changes. Using a neural circuit model of decision uncertainty and change-of-mind behaviour, we demonstrate that this phenomenon is associated with top-down signals mediated by an uncertainty-monitoring neural population. Such a mechanism is consistent with recent neurophysiological evidence showing a link between changes-of-mind and elevated top-down neural activity. Our model explains the long response times associated with changes-of-mind through high decision uncertainty levels in such trials, and accounts for the observed motor response trajectories. Overall, our work provides a computational framework that explains changes-of-mind in the absence of new post-decision evidence.

Detection of Mild Cognitive Impairment with MEG Functional Connectivity Using Wavelet-Based Neuromarkers.

Studies on developing effective neuromarkers based on magnetoencephalographic (MEG) signals have been drawing increasing attention in the neuroscience community. This study explores the idea of using source-based magnitude-squared spectral coherence as a spatial indicator for effective regions of interest (ROIs) localization, subsequently discriminating the participants with mild cognitive impairment (MCI) from a group of age-matched healthy control (HC) elderly participants. We found that the cortical regions could be divided into two distinctive groups based on their coherence indices. Compared to HC, some ROIs showed increased connectivity (hyper-connected ROIs) for MCI participants, whereas the remaining ROIs demonstrated reduced connectivity (hypo-connected ROIs). Based on these findings, a series of wavelet-based source-level neuromarkers for MCI detection are proposed and explored, with respect to the two distinctive ROI groups. It was found that the neuromarkers extracted from the hyper-connected ROIs performed significantly better for MCI detection than those from the hypo-connected ROIs. The neuromarkers were classified using support vector machine (SVM) and k-NN classifiers and evaluated through Monte Carlo cross-validation. An average recognition rate of 93.83% was obtained using source-reconstructed signals from the hyper-connected ROI group. To better conform to clinical practice settings, a leave-one-out cross-validation (LOOCV) approach was also employed to ensure that the data for testing was from a participant that the classifier has never seen. Using LOOCV, we found the best average classification accuracy was reduced to 83.80% using the same set of neuromarkers obtained from the ROI group with functional hyper-connections. This performance surpassed the results reported using wavelet-based features by approximately 15%. Overall, our work suggests that (1) certain ROIs are particularly effective for MCI detection, especially when multi-resolution wavelet biomarkers are employed for such diagnosis; (2) there exists a significant performance difference in system evaluation between research-based experimental design and clinically accepted evaluation standards.

Shaping a data-driven era in dementia care pathway through computational neurology approaches.

BACKGROUND: Dementia is caused by a variety of neurodegenerative diseases and is associated with a decline in memory and other cognitive abilities, while inflicting an enormous socioeconomic burden. The complexity of dementia and its associated comorbidities presents immense challenges for dementia research and care, particularly in clinical decision-making. MAIN BODY: Despite the lack of disease-modifying therapies, there is an increasing and urgent need to make timely and accurate clinical decisions in dementia diagnosis and prognosis to allow appropriate care and treatment. However, the dementia care pathway is currently suboptimal. We propose that through computational approaches, understanding of dementia aetiology could be improved, and dementia assessments could be more standardised, objective and efficient. In particular, we suggest that these will involve appropriate data infrastructure, the use of data-driven computational neurology approaches and the development of practical clinical decision support systems. We also discuss the technical, structural, economic, political and policy-making challenges that accompany such implementations. CONCLUSION: The data-driven era for dementia research has arrived with the potential to transform the healthcare system, creating a more efficient, transparent and personalised service for dementia.

A computational study of astrocytic glutamate influence on post-synaptic neuronal excitability.

The ability of astrocytes to rapidly clear synaptic glutamate and purposefully release the excitatory transmitter is critical in the functioning of synapses and neuronal circuits. Dysfunctions of these homeostatic functions have been implicated in the pathology of brain disorders such as mesial temporal lobe epilepsy. However, the reasons for these dysfunctions are not clear from experimental data and computational models have been developed to provide further understanding of the implications of glutamate clearance from the extracellular space, as a result of EAAT2 downregulation: although they only partially account for the glutamate clearance process. In this work, we develop an explicit model of the astrocytic glutamate transporters, providing a more complete description of the glutamate chemical potential across the astrocytic membrane and its contribution to glutamate transporter driving force based on thermodynamic principles and experimental data. Analysis of our model demonstrates that increased astrocytic glutamate content due to glutamine synthetase downregulation also results in increased postsynaptic quantal size due to gliotransmission. Moreover, the proposed model demonstrates that increased astrocytic glutamate could prolong the time course of glutamate in the synaptic cleft and enhances astrocyte-induced slow inward currents, causing a disruption to the clarity of synaptic signalling and the occurrence of intervals of higher frequency postsynaptic firing. Overall, our work distilled the necessity of a low astrocytic glutamate concentration for reliable synaptic transmission of information and the possible implications of enhanced glutamate levels as in epilepsy.

Potassium and sodium microdomains in thin astroglial processes: A computational model study.

A biophysical model that captures molecular homeostatic control of ions at the perisynaptic cradle (PsC) is of fundamental importance for understanding the interplay between astroglial and neuronal compartments. In this paper, we develop a multi-compartmental mathematical model which proposes a novel mechanism whereby the flow of cations in thin processes is restricted due to negatively charged membrane lipids which result in the formation of deep potential wells near the dipole heads. These wells restrict the flow of cations to "hopping" between adjacent wells as they transverse the process, and this surface retention of cations will be shown to give rise to the formation of potassium (K+) and sodium (Na+) microdomains at the PsC. We further propose that a K+ microdomain formed at the PsC, provides the driving force for the return of K+ to the extracellular space for uptake by the neurone, thereby preventing K+ undershoot. A slow decay of Na+ was also observed in our simulation after a period of glutamate stimulation which is in strong agreement with experimental observations. The pathological implications of microdomain formation during neuronal excitation are also discussed.

A practical computerized decision support system for predicting the severity of Alzheimer's disease of an individual.

Computerized clinical decision support systems can help to provide objective, standardized, and timely dementia diagnosis. However, current computerized systems are mainly based on group analysis, discrete classification of disease stages, or expensive and not readily accessible biomarkers, while current clinical practice relies relatively heavily on cognitive and functional assessments (CFA). In this study, we developed a computational framework using a suite of machine learning tools for identifying key markers in predicting the severity of Alzheimer's disease (AD) from a large set of biological and clinical measures. Six machine learning approaches, namely Kernel Ridge Regression (KRR), Support Vector Regression, and k-Nearest Neighbor for regression and Support Vector Machine (SVM), Random Forest, and k-Nearest Neighbor for classification, were used for the development of predictive models. We demonstrated high predictive power of CFA. Predictive performance of models incorporating CFA was shown to consistently have higher accuracy than those based solely on biomarker modalities. We found that KRR and SVM were the best performing regression and classification methods respectively. The optimal SVM performance was observed for a set of four CFA test scores (FAQ, ADAS13, MoCA, MMSE) with multi-class classification accuracy of 83.0%, 95%CI = (72.1%, 93.8%) while the best performance of the KRR model was reported with combined CFA and MRI neuroimaging data, i.e., R 2 = 0.874, 95%CI = (0.827, 0.922). Given the high predictive power of CFA and their widespread use in clinical practice, we then designed a data-driven and self-adaptive computerized clinical decision support system (CDSS) prototype for evaluating the severity of AD of an individual on a continuous spectrum. The system implemented an automated computational approach for data pre-processing, modelling, and validation and used exclusively the scores of selected cognitive measures as data entries. Taken together, we have developed an objective and practical CDSS to aid AD diagnosis.

Robust EEG/MEG Based Functional Connectivity with the Envelope of the Imaginary Coherence: Sensor Space Analysis.

The brain's functional connectivity (FC) estimated at sensor level from electromagnetic (EEG/MEG) signals can provide quick and useful information towards understanding cognition and brain disorders. Volume conduction (VC) is a fundamental issue in FC analysis due to the effects of instantaneous correlations. FC methods based on the imaginary part of the coherence (iCOH) of any two signals are readily robust to VC effects, but neglecting the real part of the coherence leads to negligible FC when the processes are truly connected but with zero or π-phase (modulus 2π) interaction. We ameliorate this issue by proposing a novel method that implements an envelope of the imaginary coherence (EIC) to approximate the coherence estimate of supposedly active underlying sources. We compare EIC with state-of-the-art FC measures that included lagged coherence, iCOH, phase lag index (PLI) and weighted PLI (wPLI), using bivariate autoregressive and stochastic neural mass models. Additionally, we create realistic simulations where three and five regions were mapped on a template cortical surface and synthetic MEG signals were obtained after computing the electromagnetic leadfield. With this simulation and comparison study, we also demonstrate the feasibility of sensor FC analysis using receiver operating curve analysis whilst varying the signal's noise level. However, these results should be interpreted with caution given the known limitations of the sensor-based FC approach. Overall, we found that EIC and iCOH demonstrate superior results with most accurate FC maps. As they complement each other in different scenarios, that will be important to study normal and diseased brain activity.

Signal Propagation in the Human Visual Pathways: An Effective Connectivity Analysis.

Although the visual system has been extensively investigated, an integrated account of the spatiotemporal dynamics of long-range signal propagation along the human visual pathways is not completely known or validated. In this work, we used dynamic causal modeling approach to provide insights into the underlying neural circuit dynamics of pattern reversal visual-evoked potentials extracted from concurrent EEG-fMRI data. A recurrent forward-backward connectivity model, consisting of multiple interacting brain regions identified by EEG source localization aided by fMRI spatial priors, best accounted for the data dynamics. Sources were first identified in the thalamic area, primary visual cortex, as well as higher cortical areas along the ventral and dorsal visual processing streams. Consistent with hierarchical early visual processing, the model disclosed and quantified the neural temporal dynamics across the identified activity sources. This signal propagation is dominated by a feedforward process, but we also found weaker effective feedback connectivity. Using effective connectivity analysis, the optimal dynamic causal modeling revealed enhanced connectivity along the dorsal pathway but slightly suppressed connectivity along the ventral pathway. A bias was also found in favor of the right hemisphere consistent with functional attentional asymmetry. This study validates, for the first time, the long-range signal propagation timing in the human visual pathways. A similar modeling approach can potentially be used to understand other cognitive processes and dysfunctions in signal propagation in neurological and neuropsychiatric disorders. Significance statement: An integrated account of long-range visual signal propagation in the human brain is currently incomplete. Using computational neural modeling on our acquired concurrent EEG-fMRI data under a visual evoked task, we found not only a substantial forward propagation toward "higher-order" brain regions but also a weaker backward propagation. Asymmetry in our model's long-range connectivity accounted for the various observed activity biases. Importantly, the model disclosed the timing of signal propagation across these connectivity pathways and validates, for the first time, long-range signal propagation in the human visual system. A similar modeling approach could be used to identify neural pathways for other cognitive processes and their dysfunctions in brain disorders.

On self-feedback connectivity in neural mass models applied to event-related potentials.

Neural mass models (NMMs) applied to neuroimaging data often do not emphasise intrinsic self-feedback within a neural population. However, based on mean-field theory, any population of coupled neurons is intrinsically endowed with effective self-coupling. In this work, we examine the effectiveness of three cortical NMMs with different self-feedbacks using a dynamic causal modelling approach. Specifically, we compare the classic Jansen and Rit (1995) model (no self-feedback), a modified model by Moran et al. (2007) (only inhibitory self-feedback), and our proposed model with inhibitory and excitatory self-feedbacks. Using bifurcation analysis, we show that single-unit Jansen-Rit model is less robust in generating oscillatory behaviour than the other two models. Next, under Bayesian inversion, we simulate single-channel event-related potentials (ERPs) within a mismatch negativity auditory oddball paradigm. We found fully self-feedback model (FSM) to provide the best fit to single-channel data. By analysing the posterior covariances of model parameters, we show that self-feedback connections are less sensitive to the generated evoked responses than the other model parameters, and hence can be treated analogously to "higher-order" parameter corrections of the original Jansen-Rit model. This is further supported in the more realistic multi-area case where FSM can replicate data better than JRM and MoM in the majority of subjects by capturing the finer features of the ERP data more accurately. Our work informs how NMMs with full self-feedback connectivity are not only more consistent with the underlying neurophysiology, but can also account for more complex features in ERP data.

Dynamic excitatory and inhibitory gain modulation can produce flexible, robust and optimal decision-making.

Behavioural and neurophysiological studies in primates have increasingly shown the involvement of urgency signals during the temporal integration of sensory evidence in perceptual decision-making. Neuronal correlates of such signals have been found in the parietal cortex, and in separate studies, demonstrated attention-induced gain modulation of both excitatory and inhibitory neurons. Although previous computational models of decision-making have incorporated gain modulation, their abstract forms do not permit an understanding of the contribution of inhibitory gain modulation. Thus, the effects of co-modulating both excitatory and inhibitory neuronal gains on decision-making dynamics and behavioural performance remain unclear. In this work, we incorporate time-dependent co-modulation of the gains of both excitatory and inhibitory neurons into our previous biologically based decision circuit model. We base our computational study in the context of two classic motion-discrimination tasks performed in animals. Our model shows that by simultaneously increasing the gains of both excitatory and inhibitory neurons, a variety of the observed dynamic neuronal firing activities can be replicated. In particular, the model can exhibit winner-take-all decision-making behaviour with higher firing rates and within a significantly more robust model parameter range. It also exhibits short-tailed reaction time distributions even when operating near a dynamical bifurcation point. The model further shows that neuronal gain modulation can compensate for weaker recurrent excitation in a decision neural circuit, and support decision formation and storage. Higher neuronal gain is also suggested in the more cognitively demanding reaction time than in the fixed delay version of the task. Using the exact temporal delays from the animal experiments, fast recruitment of gain co-modulation is shown to maximize reward rate, with a timescale that is surprisingly near the experimentally fitted value. Our work provides insights into the simultaneous and rapid modulation of excitatory and inhibitory neuronal gains, which enables flexible, robust, and optimal decision-making.

Of Rodents and Primates: Time-Variant Gain in Drift-Diffusion Decision Models.

Sequential sampling models of decision-making involve evidence accumulation over time and have been successful in capturing choice behaviour. A popular model is the drift-diffusion model (DDM). To capture the finer aspects of choice reaction times (RTs), time-variant gain features representing urgency signals have been implemented in DDM that can exhibit slower error RTs than correct RTs. However, time-variant gain is often implemented on both DDM's signal and noise features, with the assumption that increasing gain on the drift rate (due to urgency) is similar to DDM with collapsing decision bounds. Hence, it is unclear whether gain effects on just the signal or noise feature can lead to a different choice behaviour. This work presents an alternative DDM variant, focusing on the implications of time-variant gain mechanisms, constrained by model parsimony. Specifically, using computational modelling of choice behaviour of rats, monkeys, and humans, we systematically showed that time-variant gain only on the DDM's noise was sufficient to produce slower error RTs, as in monkeys, while time-variant gain only on drift rate leads to faster error RTs, as in rodents. We also found minimal effects of time-variant gain in humans. By highlighting these patterns, this study underscores the utility of group-level modelling in capturing general trends and effects consistent across species. Thus, time-variant gain on DDM's different components can lead to different choice behaviours, shed light on the underlying time-variant gain mechanisms for different species, and can be used for systematic data fitting. Supplementary Information: The online version contains supplementary material available at 10.1007/s42113-023-00194-1.

Beta-amyloid induced changes in A-type K⁺ current can alter hippocampo-septal network dynamics.

Alzheimer's disease (AD) progression is usually associated with memory deficits and cognitive decline. A hallmark of AD is the accumulation of beta-amyloid (Aß) peptide, which is known to affect the hippocampal pyramidal neurons in the early stage of AD. Previous studies have shown that Aß can block A-type K(+) currents in the hippocampal pyramidal neurons and enhance the neuronal excitability. However, the mechanisms underlying such changes and the effects of the hyper-excited pyramidal neurons on the hippocampo-septal network dynamics are still to be investigated. In this paper, Aß-blocked A-type current is simulated, and the resulting neuronal and network dynamical changes are evaluated in terms of the theta band power. The simulation results demonstrate an initial slight but significant theta band power increase as the A-type current starts to decrease. However, the theta band power eventually decreases as the A-type current is further decreased. Our analysis demonstrates that Aß blocked A-type currents can increase the pyramidal neuronal excitability by preventing the emergence of a steady state. The increased theta band power is due to more pyramidal neurons recruited into spiking mode during the peak of pyramidal theta oscillations. However, the decreased theta band power is caused by the spiking phase relationship between different neuronal populations, which is critical for theta oscillation, is violated by the hyper-excited pyramidal neurons. Our findings could provide potential implications on some AD symptoms, such as memory deficits and AD caused epilepsy.

Degeneracy and stability in neural circuits of dopamine and serotonin neuromodulators: A theoretical consideration.

Degenerate neural circuits perform the same function despite being structurally different. However, it is unclear whether neural circuits with interacting neuromodulator sources can themselves degenerate while maintaining the same neuromodulatory function. Here, we address this by computationally modeling the neural circuits of neuromodulators serotonin and dopamine, local glutamatergic and GABAergic interneurons, and their possible interactions, under reward/punishment-based conditioning tasks. The neural modeling is constrained by relevant experimental studies of the VTA or DRN system using, e.g., electrophysiology, optogenetics, and voltammetry. We first show that a single parsimonious, sparsely connected neural circuit model can recapitulate several separate experimental findings that indicated diverse, heterogeneous, distributed, and mixed DRNVTA neuronal signaling in reward and punishment tasks. The inability of this model to recapitulate all observed neuronal signaling suggests potentially multiple circuits acting in parallel. Then using computational simulations and dynamical systems analysis, we demonstrate that several different stable circuit architectures can produce the same observed network activity profile, hence demonstrating degeneracy. Due to the extensive D2-mediated connections in the investigated circuits, we simulate the D2 receptor agonist by increasing the connection strengths emanating from the VTA DA neurons. We found that the simulated D2 agonist can distinguish among sub-groups of the degenerate neural circuits based on substantial deviations in specific neural populations' activities in reward and punishment conditions. This forms a testable model prediction using pharmacological means. Overall, this theoretical work suggests the plausibility of degeneracy within neuromodulator circuitry and has important implications for the stable and robust maintenance of neuromodulatory functions.

Alzheimer's Disease Assessments Optimized for Diagnostic Accuracy and Administration Time.

OBJECTIVE: Despite the potential of machine learning techniques to improve dementia diagnostic processes, research outcomes are often not readily translated to or adopted in clinical practice. Importantly, the time taken to administer diagnostic assessment has yet to be taken into account in feature-selection based optimisation for dementia diagnosis. We address these issues by considering the impact of assessment time as a practical constraint for feature selection of cognitive and functional assessments in Alzheimer's disease diagnosis. METHODS: We use three different feature selection algorithms to select informative subsets of dementia assessment items from a large open-source dementia dataset. We use cost-sensitive feature selection to optimise our feature selection results for assessment time as well as diagnostic accuracy. To encourage clinical adoption and further evaluation of our proposed accuracy-vs-cost optimisation algorithms, we also implement a sandbox-like toolbox with graphical user interface to evaluate user-chosen subsets of assessment items. RESULTS: We find that there are subsets of accuracy-cost optimised assessment items that can perform better in terms of diagnostic accuracy and/or total assessment time than most other standard assessments. DISCUSSION: Overall, our analysis and accompanying sandbox tool can facilitate clinical users and other stakeholders to apply their own domain knowledge to analyse and decide which dementia diagnostic assessment items are useful, and aid the redesigning of dementia diagnostic assessments. Clinical Impact (Clinical Research): By optimising diagnostic accuracy and assessment time, we redesign predictive and efficient dementia diagnostic assessments and develop a sandbox interface to facilitate evaluation and testing by clinicians and non-specialists.

Alzheimer's disease classification using cluster-based labelling for graph neural network on heterogeneous data.

Biomarkers for Alzheimer's disease (AD) diagnosis do not always correlate reliably with cognitive symptoms, making clinical diagnosis inconsistent. In this study, the performance of a graphical neural network (GNN) classifier based on data-driven diagnostic classes from unsupervised clustering on heterogeneous data is compared to the performance of a classifier using clinician diagnosis as an outcome. Unsupervised clustering on tau-positron emission tomography (PET) and cognitive and functional assessment data was performed. Five clusters embedded in a non-linear uniform manifold approximation and project (UMAP) space were identified. The individual clusters revealed specific feature characteristics with respect to clinical diagnosis of AD, gender, family history, age, and underlying neurological risk factors (NRFs). In particular, one cluster comprised mainly diagnosed AD cases. All cases within this cluster were re-labelled AD cases. The re-labelled cases are characterized by high cerebrospinal fluid amyloid beta (CSF Aß) levels at a younger age, even though Aß data was not used for clustering. A GNN model was trained using the re-labelled data with a multiclass area-under-the-curve (AUC) of 95.2%, higher than the AUC of a GNN trained on clinician diagnosis (91.7%; p = 0.02). Overall, our work suggests that more objective cluster-based diagnostic labels combined with GNN classification may have value in clinical risk stratification and diagnosis of AD.

Practical Strategies for Extreme Missing Data Imputation in Dementia Diagnosis.

Accurate computational models for clinical decision support systems require clean and reliable data but, in clinical practice, data are often incomplete. Hence, missing data could arise not only from training datasets but also test datasets which could consist of a single undiagnosed case, an individual. This work addresses the problem of extreme missingness in both training and test data by evaluating multiple imputation and classification workflows based on both diagnostic classification accuracy and computational cost. Extreme missingness is defined as having ∼50% of the total data missing in more than half the data features. In particular, we focus on dementia diagnosis due to long time delays, high variability, high attrition rates and lack of practical data imputation strategies in its diagnostic pathway. We identified and replicated the extreme missingness structure of data from a real-world memory clinic on a larger open dataset, with the original complete data acting as ground truth. Overall, we found that computational cost, but not accuracy, varies widely for various imputation and classification approaches. Particularly, we found that iterative imputation on the training dataset combined with a reduced-feature classification model provides the best approach, in terms of speed and accuracy. Taken together, this work has elucidated important factors to be considered when developing a predictive model for a dementia diagnostic support system.

Multiple Cost Optimisation for Alzheimer's Disease Diagnosis.

Current machine learning techniques for dementia diagnosis often do not take into account real-world practical constraints, which may include, for example, the cost of diagnostic assessment time and financial budgets. In this work, we built on previous cost-sensitive feature selection approaches by generalising to multiple cost types, while taking into consideration that stakeholders attempting to optimise the dementia care pathway might face multiple non-fungible budget constraints. Our new optimisation algorithm involved the searching of cost-weighting hyperparameters while constrained by total budgets. We then provided a proof of concept using both assessment time cost and financial budget cost. We showed that budget constraints could control the feature selection process in an intuitive and practical manner, while adjusting the hyperparameter increased the range of solutions selected by feature selection. We further showed that our budget-constrained cost optimisation framework could be implemented in a user-friendly graphical user interface sandbox tool to encourage non-technical users and stakeholders to adopt and to further explore and audit the model - a humans-in-the-loop approach. Overall, we suggest that setting budget constraints initially and then fine tuning the cost-weighting hyperparameters can be an effective way to perform feature selection where multiple cost constraints exist, which will in turn lead to more realistic optimising and redesigning of dementia diagnostic assessments. Clinical Relevance-By optimising diagnostic accuracy against various costs (e.g. assessment administration time and financial budget) predictive yet practical dementia diagnostic assessments can be redesigned to suit clinical use.

Distinguishing Lewy Body Dementia from Alzheimer's Disease using Machine Learning on Heterogeneous Data: A Feasibility Study.

Dementia with Lewy Bodies (DLB) is the second most common form of dementia, but diagnostic markers for DLB can be expensive and inaccessible, and many cases of DLB are undiagnosed. This work applies machine learning techniques to determine the feasibility of distinguishing DLB from Alzheimer's Disease (AD) using heterogeneous data features. The Repeated Incremental Pruning to Produce Error Reduction (RIPPER) algorithm was first applied using a Leave-One-Out Cross-Validation protocol to a dataset comprising DLB and AD cases. Then, interpretable association rule-based diagnostic classifiers were obtained for distinguishing DLB from AD. The various diagnostic classifiers generated by this process had high accuracy over the whole dataset (mean accuracy of 94%). The mean accuracy in classifying their out-of-sample case was 80.5%. Every classifier generated consisted of very simple structure, each using 1-2 classification rules and 1-3 data features. As a group, the classifiers were heterogeneous and used several different data features. In particular, some of the classifiers used very simple and inexpensive diagnostic features, yet with high diagnostic accuracy. This work suggests that opportunities may exist for incorporating accessible diagnostic assessments while improving diagnostic rate for DLB. Clinical Relevance- Simple and interpretable high-performing machine learning algorithms identified a variety of readily available clinical assessments for differential diagnosis of dementia offering the opportunities to incorporate various simple and inexpensive screening tests for DLB and addressing the problem of DLB underdiagnosis.