Data-driven modelling of neurodegenerative disease progression: thinking outside the black box.

Data-driven disease progression models are an emerging set of computational tools that reconstruct disease timelines for long-term chronic diseases, providing unique insights into disease processes and their underlying mechanisms. Such methods combine a priori human knowledge and assumptions with large-scale data processing and parameter estimation to infer long-term disease trajectories from short-term data. In contrast to 'black box' machine learning tools, data-driven disease progression models typically require fewer data and are inherently interpretable, thereby aiding disease understanding in addition to enabling classification, prediction and stratification. In this Review, we place the current landscape of data-driven disease progression models in a general framework and discuss their enhanced utility for constructing a disease timeline compared with wider machine learning tools that construct static disease profiles. We review the insights they have enabled across multiple neurodegenerative diseases, notably Alzheimer disease, for applications such as determining temporal trajectories of disease biomarkers, testing hypotheses about disease mechanisms and uncovering disease subtypes. We outline key areas for technological development and translation to a broader range of neuroscience and non-neuroscience applications. Finally, we discuss potential pathways and barriers to integrating disease progression models into clinical practice and trial settings.

Towards cascading genetic risk in Alzheimer's disease.

Alzheimer's disease typically progresses in stages, which have been defined by the presence of disease-specific biomarkers: amyloid (A), tau (T) and neurodegeneration (N). This progression of biomarkers has been condensed into the ATN framework, in which each of the biomarkers can be either positive (+) or negative (-). Over the past decades, genome-wide association studies have implicated ∼90 different loci involved with the development of late-onset Alzheimer's disease. Here, we investigate whether genetic risk for Alzheimer's disease contributes equally to the progression in different disease stages or whether it exhibits a stage-dependent effect. Amyloid (A) and tau (T) status was defined using a combination of available PET and CSF biomarkers in the Alzheimer's Disease Neuroimaging Initiative cohort. In 312 participants with biomarker-confirmed A-T- status, we used Cox proportional hazards models to estimate the contribution of APOE and polygenic risk scores (beyond APOE) to convert to A+T- status (65 conversions). Furthermore, we repeated the analysis in 290 participants with A+T- status and investigated the genetic contribution to conversion to A+T+ (45 conversions). Both survival analyses were adjusted for age, sex and years of education. For progression from A-T- to A+T-, APOE-e4 burden showed a significant effect [hazard ratio (HR) = 2.88; 95% confidence interval (CI): 1.70-4.89; P < 0.001], whereas polygenic risk did not (HR = 1.09; 95% CI: 0.84-1.42; P = 0.53). Conversely, for the transition from A+T- to A+T+, the contribution of APOE-e4 burden was reduced (HR = 1.62; 95% CI: 1.05-2.51; P = 0.031), whereas the polygenic risk showed an increased contribution (HR = 1.73; 95% CI: 1.27-2.36; P < 0.001). The marginal APOE effect was driven by e4 homozygotes (HR = 2.58; 95% CI: 1.05-6.35; P = 0.039) as opposed to e4 heterozygotes (HR = 1.74; 95% CI: 0.87-3.49; P = 0.12). The genetic risk for late-onset Alzheimer's disease unfolds in a disease stage-dependent fashion. A better understanding of the interplay between disease stage and genetic risk can lead to a more mechanistic understanding of the transition between ATN stages and a better understanding of the molecular processes leading to Alzheimer's disease, in addition to opening therapeutic windows for targeted interventions.

An efficient semi-supervised quality control system trained using physics-based MRI-artefact generators and adversarial training.

Large medical imaging data sets are becoming increasingly available. A common challenge in these data sets is to ensure that each sample meets minimum quality requirements devoid of significant artefacts. Despite a wide range of existing automatic methods having been developed to identify imperfections and artefacts in medical imaging, they mostly rely on data-hungry methods. In particular, the scarcity of artefact-containing scans available for training has been a major obstacle in the development and implementation of machine learning in clinical research. To tackle this problem, we propose a novel framework having four main components: (1) a set of artefact generators inspired by magnetic resonance physics to corrupt brain MRI scans and augment a training dataset, (2) a set of abstract and engineered features to represent images compactly, (3) a feature selection process that depends on the class of artefact to improve classification performance, and (4) a set of Support Vector Machine (SVM) classifiers trained to identify artefacts. Our novel contributions are threefold: first, we use the novel physics-based artefact generators to generate synthetic brain MRI scans with controlled artefacts as a data augmentation technique. This will avoid the labour-intensive collection and labelling process of scans with rare artefacts. Second, we propose a large pool of abstract and engineered image features developed to identify 9 different artefacts for structural MRI. Finally, we use an artefact-based feature selection block that, for each class of artefacts, finds the set of features that provide the best classification performance. We performed validation experiments on a large data set of scans with artificially-generated artefacts, and in a multiple sclerosis clinical trial where real artefacts were identified by experts, showing that the proposed pipeline outperforms traditional methods. In particular, our data augmentation increases performance by up to 12.5 percentage points on the accuracy, F1, F2, precision and recall. At the same time, the computation cost of our pipeline remains low - less than a second to process a single scan - with the potential for real-time deployment. Our artefact simulators obtained using adversarial learning enable the training of a quality control system for brain MRI that otherwise would have required a much larger number of scans in both supervised and unsupervised settings. We believe that systems for quality control will enable a wide range of high-throughput clinical applications based on the use of automatic image-processing pipelines.

CenTime: Event-conditional modelling of censoring in survival analysis.

Survival analysis is a valuable tool for estimating the time until specific events, such as death or cancer recurrence, based on baseline observations. This is particularly useful in healthcare to prognostically predict clinically important events based on patient data. However, existing approaches often have limitations; some focus only on ranking patients by survivability, neglecting to estimate the actual event time, while others treat the problem as a classification task, ignoring the inherent time-ordered structure of the events. Additionally, the effective utilisation of censored samples-data points where the event time is unknown- is essential for enhancing the model's predictive accuracy. In this paper, we introduce CenTime, a novel approach to survival analysis that directly estimates the time to event. Our method features an innovative event-conditional censoring mechanism that performs robustly even when uncensored data is scarce. We demonstrate that our approach forms a consistent estimator for the event model parameters, even in the absence of uncensored data. Furthermore, CenTime is easily integrated with deep learning models with no restrictions on batch size or the number of uncensored samples. We compare our approach to standard survival analysis methods, including the Cox proportional-hazard model and DeepHit. Our results indicate that CenTime offers state-of-the-art performance in predicting time-to-death while maintaining comparable ranking performance. Our implementation is publicly available at https://github.com/ahmedhshahin/CenTime.

Multimodal subtypes identified in Alzheimer's Disease Neuroimaging Initiative participants by missing-data-enabled subtype and stage inference.

Alzheimer's disease is a highly heterogeneous disease in which different biomarkers are dynamic over different windows of the decades-long pathophysiological processes, and potentially have distinct involvement in different subgroups. Subtype and Stage Inference is an unsupervised learning algorithm that disentangles the phenotypic heterogeneity and temporal progression of disease biomarkers, providing disease insight and quantitative estimates of individual subtype and stage. However, a key limitation of Subtype and Stage Inference is that it requires a complete set of biomarkers for each subject, reducing the number of datapoints available for model fitting and limiting applications of Subtype and Stage Inference to modalities that are widely collected, e.g. volumetric biomarkers derived from structural MRI. In this study, we adapted the Subtype and Stage Inference algorithm to handle missing data, enabling the application of Subtype and Stage Inference to multimodal data (magnetic resonance imaging, positron emission tomography, cerebrospinal fluid and cognitive tests) from 789 participants in the Alzheimer's Disease Neuroimaging Initiative. Missing-data Subtype and Stage Inference identified five subtypes having distinct progression patterns, which we describe by the earliest unique abnormality as 'Typical AD with Early Tau', 'Typical AD with Late Tau', 'Cortical', 'Cognitive' and 'Subcortical'. These new multimodal subtypes were differentially associated with age, years of education, Apolipoprotein E (APOE4) status, white matter hyperintensity burden and the rate of conversion from mild cognitive impairment to Alzheimer's disease, with the 'Cognitive' subtype showing the fastest clinical progression, and the 'Subcortical' subtype the slowest. Overall, we demonstrate that missing-data Subtype and Stage Inference reveals a finer landscape of Alzheimer's disease subtypes, each of which are associated with different risk factors. Missing-data Subtype and Stage Inference has broad utility, enabling the prediction of progression in a much wider set of individuals, rather than being restricted to those with complete data.

Advanced magnetic resonance imaging detects altered placental development in pregnancies affected by congenital heart disease.

Congenital heart disease (CHD) is the most common congenital malformation and is associated with adverse neurodevelopmental outcomes. The placenta is crucial for healthy fetal development and placental development is altered in pregnancy when the fetus has CHD. This study utilized advanced combined diffusion-relaxation MRI and a data-driven analysis technique to test the hypothesis that placental microstructure and perfusion are altered in CHD-affected pregnancies. 48 participants (36 controls, 12 CHD) underwent 67 MRI scans (50 control, 17 CHD). Significant differences in the weighting of two independent placental and uterine-wall tissue components were identified between the CHD and control groups (both pFDR < 0.001), with changes most evident after 30 weeks gestation. A significant trend over gestation in weighting for a third independent tissue component was also observed in the CHD cohort (R = 0.50, pFDR = 0.04), but not in controls. These findings add to existing evidence that placental development is altered in CHD. The results may reflect alterations in placental perfusion or the changes in fetal-placental flow, villous structure and maturation that occur in CHD. Further research is needed to validate and better understand these findings and to understand the relationship between placental development, CHD, and its neurodevelopmental implications.

CF-Loss: Clinically-relevant feature optimised loss function for retinal multi-class vessel segmentation and vascular feature measurement.

Characterising clinically-relevant vascular features, such as vessel density and fractal dimension, can benefit biomarker discovery and disease diagnosis for both ophthalmic and systemic diseases. In this work, we explicitly encode vascular features into an end-to-end loss function for multi-class vessel segmentation, categorising pixels into artery, vein, uncertain pixels, and background. This clinically-relevant feature optimised loss function (CF-Loss) regulates networks to segment accurate multi-class vessel maps that produce precise vascular features. Our experiments first verify that CF-Loss significantly improves both multi-class vessel segmentation and vascular feature estimation, with two standard segmentation networks, on three publicly available datasets. We reveal that pixel-based segmentation performance is not always positively correlated with accuracy of vascular features, thus highlighting the importance of optimising vascular features directly via CF-Loss. Finally, we show that improved vascular features from CF-Loss, as biomarkers, can yield quantitative improvements in the prediction of ischaemic stroke, a real-world clinical downstream task. The code is available at https://github.com/rmaphoh/feature-loss.

Combining multimodal connectivity information improves modelling of pathology spread in Alzheimer's disease.

Cortical atrophy and aggregates of misfolded tau proteins are key hallmarks of Alzheimer's disease. Computational models that simulate the propagation of pathogens between connected brain regions have been used to elucidate mechanistic information about the spread of these disease biomarkers, such as disease epicentres and spreading rates. However, the connectomes that are used as substrates for these models are known to contain modality-specific false positive and false negative connections, influenced by the biases inherent to the different methods for estimating connections in the brain. In this work, we compare five types of connectomes for modelling both tau and atrophy patterns with the network diffusion model, which are validated against tau PET and structural MRI data from individuals with either mild cognitive impairment or dementia. We then test the hypothesis that a joint connectome, with combined information from different modalities, provides an improved substrate for the model. We find that a combination of multimodal information helps the model to capture observed patterns of tau deposition and atrophy better than any single modality. This is validated with data from independent datasets. Overall, our findings suggest that combining connectivity measures into a single connectome can mitigate some of the biases inherent to each modality and facilitate more accurate models of pathology spread, thus aiding our ability to understand disease mechanisms, and providing insight into the complementary information contained in different measures of brain connectivity.

Expectation maximisation pseudo labels.

In this paper, we study pseudo-labelling. Pseudo-labelling employs raw inferences on unlabelled data as pseudo-labels for self-training. We elucidate the empirical successes of pseudo-labelling by establishing a link between this technique and the Expectation Maximisation algorithm. Through this, we realise that the original pseudo-labelling serves as an empirical estimation of its more comprehensive underlying formulation. Following this insight, we present a full generalisation of pseudo-labels under Bayes' theorem, termed Bayesian Pseudo Labels. Subsequently, we introduce a variational approach to generate these Bayesian Pseudo Labels, involving the learning of a threshold to automatically select high-quality pseudo labels. In the remainder of the paper, we showcase the applications of pseudo-labelling and its generalised form, Bayesian Pseudo-Labelling, in the semi-supervised segmentation of medical images. Specifically, we focus on: (1) 3D binary segmentation of lung vessels from CT volumes; (2) 2D multi-class segmentation of brain tumours from MRI volumes; (3) 3D binary segmentation of whole brain tumours from MRI volumes; and (4) 3D binary segmentation of prostate from MRI volumes. We further demonstrate that pseudo-labels can enhance the robustness of the learned representations. The code is released in the following GitHub repository: https://github.com/moucheng2017/EMSSL.

The Variegation of Human Brain Vulnerability to Rare Genetic Disorders and Convergence With Behaviorally Defined Disorders.

BACKGROUND: Diverse gene dosage disorders (GDDs) increase risk for psychiatric impairment, but characterization of GDD effects on the human brain has so far been piecemeal, with few simultaneous analyses of multiple brain features across different GDDs. METHODS: Here, through multimodal neuroimaging of 3 aneuploidy syndromes (XXY [total n = 191, 92 control participants], XYY [total n = 81, 47 control participants], and trisomy 21 [total n = 69, 41 control participants]), we systematically mapped the effects of supernumerary X, Y, and chromosome 21 dosage across a breadth of 15 different macrostructural, microstructural, and functional imaging-derived phenotypes (IDPs). RESULTS: The results revealed considerable diversity in cortical changes across GDDs and IDPs. This variegation of IDP change underlines the limitations of studying GDD effects unimodally. Integration across all IDP change maps revealed highly distinct architectures of cortical change in each GDD along with partial coalescence onto a common spatial axis of cortical vulnerability that is evident in all 3 GDDs. This common axis shows strong alignment with shared cortical changes in behaviorally defined psychiatric disorders and is enriched for specific molecular and cellular signatures. CONCLUSIONS: Use of multimodal neuroimaging data in 3 aneuploidies indicates that different GDDs impose unique fingerprints of change in the human brain that differ widely depending on the imaging modality that is being considered. Embedded in this variegation is a spatial axis of shared multimodal change that aligns with shared brain changes across psychiatric disorders and therefore represents a major high-priority target for future translational research in neuroscience.

Enhancing cancer prediction in challenging screen-detected incident lung nodules using time-series deep learning.

Lung cancer screening (LCS) using annual computed tomography (CT) scanning significantly reduces mortality by detecting cancerous lung nodules at an earlier stage. Deep learning algorithms can improve nodule malignancy risk stratification. However, they have typically been used to analyse single time point CT data when detecting malignant nodules on either baseline or incident CT LCS rounds. Deep learning algorithms have the greatest value in two aspects. These approaches have great potential in assessing nodule change across time-series CT scans where subtle changes may be challenging to identify using the human eye alone. Moreover, they could be targeted to detect nodules developing on incident screening rounds, where cancers are generally smaller and more challenging to detect confidently. Here, we show the performance of our Deep learning-based Computer-Aided Diagnosis model integrating Nodule and Lung imaging data with clinical Metadata Longitudinally (DeepCAD-NLM-L) for malignancy prediction. DeepCAD-NLM-L showed improved performance (AUC = 88%) against models utilizing single time-point data alone. DeepCAD-NLM-L also demonstrated comparable and complementary performance to radiologists when interpreting the most challenging nodules typically found in LCS programs. It also demonstrated similar performance to radiologists when assessed on out-of-distribution imaging dataset. The results emphasize the advantages of using time-series and multimodal analyses when interpreting malignancy risk in LCS.

Interpolation-split: a data-centric deep learning approach with big interpolated data to boost airway segmentation performance.

The morphology and distribution of airway tree abnormalities enable diagnosis and disease characterisation across a variety of chronic respiratory conditions. In this regard, airway segmentation plays a critical role in the production of the outline of the entire airway tree to enable estimation of disease extent and severity. Furthermore, the segmentation of a complete airway tree is challenging as the intensity, scale/size and shape of airway segments and their walls change across generations. The existing classical techniques either provide an undersegmented or oversegmented airway tree, and manual intervention is required for optimal airway tree segmentation. The recent development of deep learning methods provides a fully automatic way of segmenting airway trees; however, these methods usually require high GPU memory usage and are difficult to implement in low computational resource environments. Therefore, in this study, we propose a data-centric deep learning technique with big interpolated data, Interpolation-Split, to boost the segmentation performance of the airway tree. The proposed technique utilises interpolation and image split to improve data usefulness and quality. Then, an ensemble learning strategy is implemented to aggregate the segmented airway segments at different scales. In terms of average segmentation performance (dice similarity coefficient, DSC), our method (A) achieves 90.55%, 89.52%, and 85.80%; (B) outperforms the baseline models by 2.89%, 3.86%, and 3.87% on average; and (C) produces maximum segmentation performance gain by 14.11%, 9.28%, and 12.70% for individual cases when (1) nnU-Net with instant normalisation and leaky ReLU; (2) nnU-Net with batch normalisation and ReLU; and (3) modified dilated U-Net are used respectively. Our proposed method outperformed the state-of-the-art airway segmentation approaches. Furthermore, our proposed technique has low RAM and GPU memory usage, and it is GPU memory-efficient and highly flexible, enabling it to be deployed on any 2D deep learning model.

Longitudinal effects of prenatal alcohol exposure on visual neurodevelopment over infancy.

Prenatal alcohol exposure (PAE) affects neurodevelopment in over 59 million individuals globally. Prior studies using dichotomous categorization of alcohol use and comorbid substance exposures provide limited knowledge of how prenatal alcohol specifically impacts early human neurodevelopment. In this longitudinal cohort study from Cape Town, South Africa, PAE is measured continuously-characterizing timing, dose, and drinking patterns (i.e., binge drinking). High-density electroencephalography (EEG) during a visual-evoked potential (VEP) task was collected from infants aged 8 to 52 weeks with prenatal exposure exclusively to alcohol and matched on sociodemographic factors to infants with no substance exposure in utero. First trimester alcohol exposure related to altered timing of the P1 VEP component over the first 6 months postnatally, and first trimester binge drinking exposure altered timing of the P1 VEP components such that increased exposure was associated with longer VEP latencies while increasing age was related to shorter VEP latencies (n = 108). These results suggest alcohol exposure in the first trimester may alter visual neurodevelopmental timing in early infancy. Exploratory individual-difference analysis across infants with and without PAE tested the relation between VEP latencies and myelination for a subsample of infants with usable magnetic resonance imaging (MRI) T1w and T2w scans collected at the same time point as EEG (n = 47). Decreased MRI T1w/T2w ratios (an indicator of myelin) in the primary visual cortex (n = 47) were linked to longer P1 VEP latencies. Results from these two sets of analyses suggest that prenatal alcohol and postnatal myelination may both separately impact VEP latency over infancy. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Homopolymer switches mediate adaptive mutability in mismatch repair-deficient colorectal cancer.

Mismatch repair (MMR)-deficient cancer evolves through the stepwise erosion of coding homopolymers in target genes. Curiously, the MMR genes MutS homolog 6 (MSH6) and MutS homolog 3 (MSH3) also contain coding homopolymers, and these are frequent mutational targets in MMR-deficient cancers. The impact of incremental MMR mutations on MMR-deficient cancer evolution is unknown. Here we show that microsatellite instability modulates DNA repair by toggling hypermutable mononucleotide homopolymer runs in MSH6 and MSH3 through stochastic frameshift switching. Spontaneous mutation and reversion modulate subclonal mutation rate, mutation bias and HLA and neoantigen diversity. Patient-derived organoids corroborate these observations and show that MMR homopolymer sequences drift back into reading frame in the absence of immune selection, suggesting a fitness cost of elevated mutation rates. Combined experimental and simulation studies demonstrate that subclonal immune selection favors incremental MMR mutations. Overall, our data demonstrate that MMR-deficient colorectal cancers fuel intratumor heterogeneity by adapting subclonal mutation rate and diversity to immune selection.