Can machine learning of post-procedural cone-beam CT images in acute ischemic stroke improve the detection of 24-h hemorrhagic transformation? A preliminary study.

PURPOSE: Hemorrhagic transformation (HT) is an independent predictor of unfavorable outcome in acute ischemic stroke (AIS) patients undergoing endovascular thrombectomy (EVT). Its early identification could help tailor AIS management. We hypothesize that machine learning (ML) applied to cone-beam computed tomography (CB-CT), immediately after EVT, improves performance in 24-h HT prediction. METHODS: We prospectively enrolled AIS patients undergoing EVT, post-procedural CB-CT, and 24-h non-contrast CT (NCCT). Three raters independently analyzed imaging at four anatomic levels qualitatively and quantitatively selecting a region of interest (ROI) < 5 mm2. Each ROI was labeled as "hemorrhagic" or "non-hemorrhagic" depending on 24-h NCCT. For each level of CB-CT, Mean Hounsfield Unit (HU), minimum HU, maximum HU, and signal- and contrast-to-noise ratios were calculated, and the differential HU-ROI value was compared between both hemispheres. The number of anatomic levels affected was computed for lesion volume estimation. ML with the best validation performance for 24-h HT prediction was selected. RESULTS: One hundred seventy-two ROIs from affected hemispheres of 43 patients were extracted. Ninety-two ROIs were classified as unremarkable, whereas 5 as parenchymal contrast staining, 29 as ischemia, 7 as subarachnoid hemorrhages, and 39 as HT. The Bernoulli Naïve Bayes was the best ML classifier with a good performance for 24-h HT prediction (sensitivity = 1.00; specificity = 0.75; accuracy = 0.82), though precision was 0.60. CONCLUSION: ML demonstrates high-sensitivity but low-accuracy 24-h HT prediction in AIS. The automated CB-CT imaging evaluation resizes sensitivity, specificity, and accuracy rates of visual interpretation reported in the literature so far. A standardized quantitative interpretation of CB-CT may be warranted to overcome the inter-operator variability.

Deep computational pathology in breast cancer.

Deep Learning (DL) algorithms are a set of techniques that exploit large and/or complex real-world datasets for cross-domain and cross-discipline prediction and classification tasks. DL architectures excel in computer vision tasks, and in particular image processing and interpretation. This has prompted a wave of disruptingly innovative applications in medical imaging, where DL strategies have the potential to vastly outperform human experts. This is particularly relevant in the context of histopathology, where whole slide imaging (WSI) of stained tissue in conjuction with DL algorithms for their interpretation, selection and cancer staging are beginning to play an ever increasing role in supporting human operators in visual assessments. This has the potential to reduce everyday workload as well as to increase precision and reproducibility across observers, centers, staining techniques and even pathologies. In this paper we introduce the most common DL architectures used in image analysis, with a focus on histopathological image analysis in general and in breast histology in particular. We briefly review how, state-of-art DL architectures compare to human performance on across a number of critical tasks such as mitotic count, tubules analysis and nuclear pleomorphism analysis. Also, the development of DL algorithms specialized to pathology images have been enormously fueled by a number of world-wide challenges based on large, multicentric image databases which are now publicly available. In turn, this has allowed most recent efforts to shift more and more towards semi-supervised learning methods, which provide greater flexibility and applicability. We also review all major repositories of manually labelled pathology images in breast cancer and provide an in-depth discussion of the challenges specific to training DL architectures to interpret WSI data, as well as a review of the state-of-the-art methods for interpretation of images generated from immunohistochemical analysis of breast lesions. We finally discuss the future challenges and opportunities which the adoption of DL paradigms is most likely to pose in the field of pathology for breast cancer detection, diagnosis, staging and prognosis. This review is intended as a comprehensive stepping stone into the field of modern computational pathology for a transdisciplinary readership across technical and medical disciplines.

Radiomics in breast cancer classification and prediction.

Breast Cancer (BC) is the common form of cancer in women. Its diagnosis and screening are usually performed through different imaging modalities such as mammography, magnetic resonance imaging and ultrasound. However, mammography and ultrasound-imaging techniques have limited sensitivity and specificity both in identifying lesions and in differentiating malign from benign lesions, especially in presence of dense breast parenchyma. Due to the higher resolution of magnetic resonance images, MRI represents the method with the higher specificity and sensitivity among all the available tools, in both lesions' identification and diagnosis. However, especially for diagnosis, even MRI has limitations that are only partially solved if combined with mammography. Unfortunately, due to the limits of all these imaging tools, in order to have a certain diagnosis, patients often receive painful and costly bioptics procedures. In this context, several computational approaches have been developed to increase sensitivity, while maintaining the same specificity, in BC diagnosis and screening. Amongst these, radiomics has been increasingly gaining ground in oncology to improve cancer diagnosis, prognosis and treatment. Radiomics derives multiple quantitative features from single or multiple medical imaging modalities, highlighting image traits which are not visible to the naked eye and hence significantly augmenting the discriminatory and predictive potential of medical imaging. This review article aims to summarize the state of the art in radiomics-based BC research. The dominating evidence extracted from the literature points towards a high potential of radiomics in disentangling malignant from benign breast lesions, classifying BC types and grades and also in predicting treatment response and recurrence risk. In the era of personalized medicine, radiomics has the potential to improve diagnosis, prognosis, prediction, monitoring, image-based intervention, and assessment of therapeutic response in BC.

Echo state network models for nonlinear Granger causality.

While Granger causality (GC) has been often employed in network neuroscience, most GC applications are based on linear multivariate autoregressive (MVAR) models. However, real-life systems like biological networks exhibit notable nonlinear behaviour, hence undermining the validity of MVAR-based GC (MVAR-GC). Most nonlinear GC estimators only cater for additive nonlinearities or, alternatively, are based on recurrent neural networks or long short-term memory networks, which present considerable training difficulties and tailoring needs. We reformulate the GC framework in terms of echo-state networks-based models for arbitrarily complex networks, and characterize its ability to capture nonlinear causal relations in a network of noisy Duffing oscillators, showing a net advantage of echo state GC (ES-GC) in detecting nonlinear, causal links. We then explore the structure of ES-GC networks in the human brain employing functional MRI data from 1003 healthy subjects drawn from the human connectome project, demonstrating the existence of previously unknown directed within-brain interactions. In addition, we examine joint brain-heart signals in 15 subjects where we explore directed interaction between brain networks and central vagal cardiac control in order to investigate the so-called central autonomic network in a causal manner. This article is part of the theme issue 'Advanced computation in cardiovascular physiology: new challenges and opportunities'.

A novel multi-branch architecture for state of the art robust detection of pathological phonocardiograms.

Heart auscultation is an inexpensive and fundamental technique to effectively diagnose cardiovascular disease. However, due to relatively high human error rates even when auscultation is performed by an experienced physician, and due to the not universal availability of qualified personnel, e.g. in developing countries, many efforts are made worldwide to propose computational tools for detecting abnormalities in heart sounds. The large heterogeneity of achievable data quality and devices, the variety of possible heart pathologies, and a generally poor signal-to-noise ratio make this problem very challenging. We present an accurate classification strategy for diagnosing heart sounds based on (1) automatic heart phase segmentation, (2) state-of-the art filters drawn from the field of speech synthesis (mel-frequency cepstral representation) and (3) an ad hoc multi-branch, multi-instance artificial neural network based on convolutional layers and fully connected neuronal ensembles which separately learns from each heart phase hence implicitly leveraging their different physiological significance. We demonstrate that it is possible to train our architecture to reach very high performances, e.g. an area under the curve of 0.87 or a sensitivity of 0.97. Our machine-learning-based tool could be employed for heartsound classification, especially as a screening tool in a variety of situations including telemedicine applications. This article is part of the theme issue 'Advanced computation in cardiovascular physiology: new challenges and opportunities'.

A parameter-efficient deep learning approach to predict conversion from mild cognitive impairment to Alzheimer's disease.

Some forms of mild cognitive impairment (MCI) are the clinical precursors of Alzheimer's disease (AD), while other MCI types tend to remain stable over-time and do not progress to AD. To identify and choose effective and personalized strategies to prevent or slow the progression of AD, we need to develop objective measures that are able to discriminate the MCI patients who are at risk of AD from those MCI patients who have less risk to develop AD. Here, we present a novel deep learning architecture, based on dual learning and an ad hoc layer for 3D separable convolutions, which aims at identifying MCI patients who have a high likelihood of developing AD within 3 years. Our deep learning procedures combine structural magnetic resonance imaging (MRI), demographic, neuropsychological, and APOe4 genetic data as input measures. The most novel characteristics of our machine learning model compared to previous ones are the following: 1) our deep learning model is multi-tasking, in the sense that it jointly learns to simultaneously predict both MCI to AD conversion as well as AD vs. healthy controls classification, which facilitates relevant feature extraction for AD prognostication; 2) the neural network classifier employs fewer parameters than other deep learning architectures which significantly limits data-overfitting (we use ∼550,000 network parameters, which is orders of magnitude lower than other network designs); 3) both structural MRI images and their warp field characteristics, which quantify local volumetric changes in relation to the MRI template, were used as separate input streams to extract as much information as possible from the MRI data. All analyses were performed on a subset of the database made publicly available via the Alzheimer's Disease Neuroimaging Initiative (ADNI), (n = 785 participants, n = 192 AD patients, n = 409 MCI patients (including both MCI patients who convert to AD and MCI patients who do not covert to AD), and n = 184 healthy controls). The most predictive combination of inputs were the structural MRI images and the demographic, neuropsychological, and APOe4 data. In contrast, the warp field metrics were of little added predictive value. The algorithm was able to distinguish the MCI patients developing AD within 3 years from those patients with stable MCI over the same time-period with an area under the curve (AUC) of 0.925 and a 10-fold cross-validated accuracy of 86%, a sensitivity of 87.5%, and specificity of 85%. To our knowledge, this is the highest performance achieved so far using similar datasets. The same network provided an AUC of 1 and 100% accuracy, sensitivity, and specificity when classifying patients with AD from healthy controls. Our classification framework was also robust to the use of different co-registration templates and potentially irrelevant features/image portions. Our approach is flexible and can in principle integrate other imaging modalities, such as PET, and diverse other sets of clinical data. The convolutional framework is potentially applicable to any 3D image dataset and gives the flexibility to design a computer-aided diagnosis system targeting the prediction of several medical conditions and neuropsychiatric disorders via multi-modal imaging and tabular clinical data.

Variability and Reproducibility of Directed and Undirected Functional MRI Connectomes in the Human Brain.

A growing number of studies are focusing on methods to estimate and analyze the functional connectome of the human brain. Graph theoretical measures are commonly employed to interpret and synthesize complex network-related information. While resting state functional MRI (rsfMRI) is often employed in this context, it is known to exhibit poor reproducibility, a key factor which is commonly neglected in typical cohort studies using connectomics-related measures as biomarkers. We aimed to fill this gap by analyzing and comparing the inter- and intra-subject variability of connectivity matrices, as well as graph-theoretical measures, in a large (n = 1003) database of young healthy subjects which underwent four consecutive rsfMRI sessions. We analyzed both directed (Granger Causality and Transfer Entropy) and undirected (Pearson Correlation and Partial Correlation) time-series association measures and related global and local graph-theoretical measures. While matrix weights exhibit a higher reproducibility in undirected, as opposed to directed, methods, this difference disappears when looking at global graph metrics and, in turn, exhibits strong regional dependence in local graphs metrics. Our results warrant caution in the interpretation of connectivity studies, and serve as a benchmark for future investigations by providing quantitative estimates for the inter- and intra-subject variabilities in both directed and undirected connectomic measures.

A Parsimonious Granger Causality Formulation for Capturing Arbitrarily Long Multivariate Associations.

High-frequency neuroelectric signals like electroencephalography (EEG) or magnetoencephalography (MEG) provide a unique opportunity to infer causal relationships between local activity of brain areas. While causal inference is commonly performed through classical Granger causality (GC) based on multivariate autoregressive models, this method may encounter important limitations (e.g., data paucity) in the case of high dimensional data from densely connected systems like the brain. Additionally, physiological signals often present long-range dependencies which commonly require high autoregressive model orders/number of parameters. We present a generalization of autoregressive models for GC estimation based on Wiener-Volterra decompositions with Laguerre polynomials as basis functions. In this basis, the introduction of only one additional global parameter allows to capture arbitrary long dependencies without increasing model order, hence retaining model simplicity, linearity and ease of parameters estimation. We validate our method in synthetic data generated from families of complex, densely connected networks and demonstrate superior performance as compared to classical GC. Additionally, we apply our framework to studying the directed human brain connectome through MEG data from 89 subjects drawn from the Human Connectome Project (HCP) database, showing that it is able to reproduce current knowledge as well as to uncover previously unknown directed influences between cortical and limbic brain regions.

Functional connectivity in amygdalar-sensory/(pre)motor networks at rest: new evidence from the Human Connectome Project.

The word 'e-motion' derives from the Latin word 'ex-moveo' which literally means 'moving away from something/somebody'. Emotions are thus fundamental to prime action and goal-directed behavior with obvious implications for individual's survival. However, the brain mechanisms underlying the interactions between emotional and motor cortical systems remain poorly understood. A recent diffusion tensor imaging study in humans has reported the existence of direct anatomical connections between the amygdala and sensory/(pre)motor cortices, corroborating an initial observation in animal research. Nevertheless, the functional significance of these amygdala-sensory/(pre)motor pathways remain uncertain. More specifically, it is currently unclear whether a distinct amygdala-sensory/(pre)motor circuit can be identified with resting-state functional magnetic resonance imaging (rs-fMRI). This is a key issue, as rs-fMRI offers an opportunity to simultaneously examine distinct neural circuits that underpin different cognitive, emotional and motor functions, while minimizing task-related performance confounds. We therefore tested the hypothesis that the amygdala and sensory/(pre)motor cortices could be identified as part of the same resting-state functional connectivity network. To this end, we examined independent component analysis results in a very large rs-fMRI data-set drawn from the Human Connectome Project (n = 820 participants, mean age: 28.5 years). To our knowledge, we report for the first time the existence of a distinct amygdala-sensory/(pre)motor functional network at rest. rs-fMRI studies are now warranted to examine potential abnormalities in this circuit in psychiatric and neurological diseases that may be associated with alterations in the amygdala-sensory/(pre)motor pathways (e.g. conversion disorders, impulse control disorders, amyotrophic lateral sclerosis and multiple sclerosis).

Heart rate variability in untreated newly diagnosed temporal lobe epilepsy: Evidence for ictal sympathetic dysregulation.

OBJECTIVE: To compare heart rate variability (HRV) parameters in newly diagnosed and untreated temporal lobe epilepsy (TLE) between the interictal, preictal, ictal, and postictal states. METHODS: HRV parameters were extracted from single-lead electrocardiography data collected during video-electroencephalography (EEG) recordings from 14 patients with newly diagnosed TLE in a resting, awake, and supine state. HRV parameters in the time and frequency domains included low frequency (LF), high frequency (HF), standard deviation of all consecutive R wave intervals (SDNN), and square root of the mean of the sum of the squares of differences between adjacent R wave intervals (RMSSD). Cardiovagal index (CVI), cardiosympathetic index (CSI), and approximate entropy (ApEn) were also studied. RESULTS: Frequency domain analysis showed significantly higher preictal, ictal, and postictal LF/HF ratio compared to the interictal state. Similarly, the LF component increased progressively and was significantly higher during the ictal state compared to interictal and preictal states. RR interval values were lower in the ictal state compared to basal and preictal states and in the postictal state compared to the preictal state. Interictal RMSSD was significantly higher compared to all other states, and ictal SDNN was significantly higher compared to all other states. Ictal CSI was significantly higher compared to preictal and interictal states, whereas preictal CVI was lower than in basal and ictal states. In addition, ictal ApEn was significantly lower than interictal and preictal ApEn. Interictal CVI was lower in left TLE compared to right TLE. In addition, in left TLE, ictal CVI was higher than interictal CVI, whereas in right TLE, CVI was lower in the preictal state compared to all other states. SIGNIFICANCE: Our data suggest an ictal sympathetic overdrive with partial recovery in the postictal state. Higher sympathetic tone and vagal tone imbalance may induce early autonomic dysfunction and increase cardiovascular risk in patients affected by TLE.

Beyond multilayer perceptrons: Investigating complex topologies in neural networks.

This study delves into the crucial aspect of network topology in artificial neural networks (NNs) and its impact on model performance. Addressing the need to comprehend how network structures influence learning capabilities, the research contrasts traditional multilayer perceptrons (MLPs) with models built on various complex topologies using novel network generation techniques. Drawing insights from synthetic datasets, the study reveals the remarkable accuracy of complex NNs, particularly in high-difficulty scenarios, outperforming MLPs. Our exploration extends to real-world datasets, highlighting the task-specific nature of optimal network topologies and unveiling trade-offs, including increased computational demands and reduced robustness to graph damage in complex NNs compared to MLPs. This research underscores the pivotal role of complex topologies in addressing challenging learning tasks. However, it also signals the necessity for deeper insights into the complex interplay among topological attributes influencing NN performance. By shedding light on the advantages and limitations of complex topologies, this study provides valuable guidance for practitioners and paves the way for future endeavors to design more efficient and adaptable neural architectures across various applications.

Heritability of human "directed" functional connectome.

INTRODUCTION: The functional connectivity patterns in the brain are highly heritable; however, it is unclear how genetic factors influence the directionality of such "information flows." Studying the "directionality" of the brain functional connectivity and assessing how heritability modulates it can improve our understanding of the human connectome. METHODS: Here, we investigated the heritability of "directed" functional connections using a state-space formulation of Granger causality (GC), in conjunction with blind deconvolution methods accounting for local variability in the hemodynamic response function. Such GC implementation is ideal to explore the directionality of functional interactions across a large number of networks. Resting-state functional magnetic resonance imaging data were drawn from the Human Connectome Project (total n = 898 participants). To add robustness to our findings, the dataset was randomly split into a "discovery" and a "replication" sample (each with n = 449 participants). The two cohorts were carefully matched in terms of demographic variables and other confounding factors (e.g., education). The effect of shared environment was also modeled. RESULTS: The parieto- and prefronto-cerebellar, parieto-prefrontal, and posterior-cingulate to hippocampus connections showed the highest and most replicable heritability effects with little influence by shared environment. In contrast, shared environmental factors significantly affected the visuo-parietal and sensory-motor directed connectivity. CONCLUSION: We suggest a robust role of heritability in influencing the directed connectivity of some cortico-subcortical circuits implicated in cognition. Further studies, for example using task-based fMRI and GC, are warranted to confirm the asymmetric effects of genetic factors on the functional connectivity within cognitive networks and their role in supporting executive functions and learning.

Causal influence of brainstem response to transcutaneous vagus nerve stimulation on cardiovagal outflow.

BACKGROUND: The autonomic response to transcutaneous auricular vagus nerve stimulation (taVNS) has been linked to the engagement of brainstem circuitry modulating autonomic outflow. However, the physiological mechanisms supporting such efferent vagal responses are not well understood, particularly in humans. HYPOTHESIS: We present a paradigm for estimating directional brain-heart interactions in response to taVNS. We propose that our approach is able to identify causal links between the activity of brainstem nuclei involved in autonomic control and cardiovagal outflow. METHODS: We adopt an approach based on a recent reformulation of Granger causality that includes permutation-based, nonparametric statistics. The method is applied to ultrahigh field (7T) functional magnetic resonance imaging (fMRI) data collected on healthy subjects during taVNS. RESULTS: Our framework identified taVNS-evoked functional brainstem responses with superior sensitivity compared to prior conventional approaches, confirming causal links between taVNS stimulation and fMRI response in the nucleus tractus solitarii (NTS). Furthermore, our causal approach elucidated potential mechanisms by which information is relayed between brainstem nuclei and cardiovagal, i.e., high-frequency heart rate variability, in response to taVNS. Our findings revealed that key brainstem nuclei, known from animal models to be involved in cardiovascular control, exert a causal influence on taVNS-induced cardiovagal outflow in humans. CONCLUSION: Our causal approach allowed us to noninvasively evaluate directional interactions between fMRI BOLD signals from brainstem nuclei and cardiovagal outflow.

Inference of time-evolving coupled dynamical systems in the presence of noise.

A new method is introduced for analysis of interactions between time-dependent coupled oscillators, based on the signals they generate. It distinguishes unsynchronized dynamics from noise-induced phase slips and enables the evolution of the coupling functions and other parameters to be followed. It is based on phase dynamics, with Bayesian inference of the time-evolving parameters achieved by shaping the prior densities to incorporate knowledge of previous samples. The method is tested numerically and applied to reveal and quantify the time-varying nature of cardiorespiratory interactions.

Physically constrained neural networks for inferring physiological system models.

Systems biology and systems neurophysiology in particular have recently emerged as powerful tools for a number of key applications in the biomedical sciences. Nevertheless, such models are often based on complex combinations of multiscale (and possibly multiphysics) strategies that require ad hoc computational strategies and pose extremely high computational demands. Recent developments in the field of deep neural networks have demonstrated the possibility of formulating nonlinear, universal approximators to estimate solutions to highly nonlinear and complex problems with significant speed and accuracy advantages in comparison with traditional models. After synthetic data validation, we use so-called physically constrained neural networks (PINN) to simultaneously solve the biologically plausible Hodgkin-Huxley model and infer its parameters and hidden time-courses from real data under both variable and constant current stimulation, demonstrating extremely low variability across spikes and faithful signal reconstruction. The parameter ranges we obtain are also compatible with prior knowledge. We demonstrate that detailed biological knowledge can be provided to a neural network, making it able to fit complex dynamics over both simulated and real data.

Spatiotemporal learning of dynamic positron emission tomography data improves diagnostic accuracy in breast cancer.

Positron emission tomography (PET) can reveal metabolic activity in a voxelwise manner. PET analysis is commonly performed in a static manner by analyzing the standardized uptake value (SUV) obtained from the plateau region of PET acquisitions. A dynamic PET acquisition can provide a map of the spatiotemporal concentration of the tracer in vivo, hence conveying information about radiotracer delivery to tissue, its interaction with the target and washout. Therefore, tissue-specific biochemical properties are embedded in the shape of time activity curves (TACs), which are generally used for kinetic analysis. Conventionally, TACs are employed along with information about blood plasma activity concentration, i.e., the arterial input function (AIF), and specific compartmental models to obtain a full quantitative analysis of PET data. The main drawback of this approach is the need for invasive procedures requiring arterial blood sample collection during the whole PET scan. In this paper, we address the challenge of improving PET diagnostic accuracy through an alternative approach based on the analysis of time signal intensity patterns. Specifically, we demonstrate the diagnostic potential of tissue TACs provided by dynamic PET acquisition using various deep learning models. Our framework is shown to outperform the discriminative potential of classical SUV analysis, hence paving the way for more accurate PET-based lesion discrimination without additional acquisition time or invasive procedures. Clinical Relevance- The diagnostic accuracy of dynamic PET data exploited by deep-learning based time signal intensity pattern analysis is superior to that of static SUV imaging.

Patient-clinician brain concordance underlies causal dynamics in nonverbal communication and negative affective expressivity.

Patient-clinician concordance in behavior and brain activity has been proposed as a potential key mediator of mutual empathy and clinical rapport in the therapeutic encounter. However, the specific elements of patient-clinician communication that may support brain-to-brain concordance and therapeutic alliance are unknown. Here, we investigated how pain-related, directional facial communication between patients and clinicians is associated with brain-to-brain concordance. Patient-clinician dyads interacted in a pain-treatment context, during synchronous assessment of brain activity (fMRI hyperscanning) and online video transfer, enabling face-to-face social interaction. In-scanner videos were used for automated individual facial action unit (AU) time-series extraction. First, an interpretable machine-learning classifier of patients' facial expressions, from an independent fMRI experiment, significantly distinguished moderately painful leg pressure from innocuous pressure stimuli. Next, we estimated neural-network causality of patient-to-clinician directional information flow of facial expressions during clinician-initiated treatment of patients' evoked pain. We identified a leader-follower relationship in which patients predominantly led the facial communication while clinicians responded to patients' expressions. Finally, analyses of dynamic brain-to-brain concordance showed that patients' mid/posterior insular concordance with the clinicians' anterior insula cortex, a region identified in previously published data from this study1, was associated with therapeutic alliance, and self-reported and objective (patient-to-clinician-directed causal influence) markers of negative-affect expressivity. These results suggest a role of patient-clinician concordance of the insula, a social-mirroring and salience-processing brain node, in mediating directional dynamics of pain-directed facial communication during therapeutic encounters.

Classification of real-world pathological phonocardiograms through multi-instance learning.

Heart auscultation is an inexpensive and fundamental technique to effectively to diagnose cardiovascular disease. However, due to relatively high human error rates even when auscultation is performed by an experienced physician, and due to the not universal availability of qualified personnel e.g. in developing countries, a large body of research is attempting to develop automated, computational tools for detecting abnormalities in heart sounds. The large heterogeneity of achievable data quality and devices, the variety o possible heart pathologies, and a generally poor signal-to-noise ratio make this problem extremely challenging. We present an accurate classification strategy for diagnosing heart sounds based on 1) automatic heart phase segmentation, 2) state-of-the art filters drawn from the filed of speech synthesis (mel-frequency cepstral representation), and 3) an ad-hoc multi-branch, multi-instance artificial neural network based on convolutional layers and fully connected neuronal ensembles which separately learns from each heart phase, hence leveraging their different physiological significance. We demonstrate that it is possible to train our architecture to reach very high performances, e.g. an AUC of 0.87 or a sensitivity of 0.97. Our machine-learning-based tool could be employed for heart sound classification, especially as a screening tool in a variety of situations including telemedicine applications.

Thalamic neuroinflammation as a reproducible and discriminating signature for chronic low back pain.

ABSTRACT: Using positron emission tomography, we recently demonstrated elevated brain levels of the 18 kDa translocator protein (TSPO), a glial activation marker, in chronic low back pain (cLBP) patients, compared to healthy controls (HCs). Here, we first sought to replicate the original findings in an independent cohort (15 cLBP, 37.8 ± 12.5 y/o; 18 HC, 48.2 ± 12.8 y/o). We then trained random forest machine learning algorithms based on TSPO imaging features combining discovery and replication cohorts (totaling 25 cLBP, 42.4 ± 13.2 y/o; 27 HC, 48.9 ± 12.6 y/o), to explore whether image features other than the mean contain meaningful information that might contribute to the discrimination of cLBP patients and HC. Feature importance was ranked using SHapley Additive exPlanations values, and the classification performance (in terms of area under the curve values) of classifiers containing only the mean, other features, or all features was compared using the DeLong test. Both region-of-interest and voxelwise analyses replicated the original observation of thalamic TSPO signal elevations in cLBP patients compared to HC (P < 0.05). The random forest-based analyses revealed that although the mean is a discriminating feature, other features demonstrate similar level of importance, including the maximum, kurtosis, and entropy. Our observations suggest that thalamic neuroinflammatory signal is a reproducible and discriminating feature for cLBP, further supporting a role for glial activation in human cLBP, and the exploration of neuroinflammation as a therapeutic target for chronic pain. This work further shows that TSPO signal contains a richness of information that the simple mean might fail to capture completely.

Uncovering complex central autonomic networks at rest: a functional magnetic resonance imaging study on complex cardiovascular oscillations.

This study aims to uncover brain areas that are functionally linked to complex cardiovascular oscillations in resting-state conditions. Multi-session functional magnetic resonance imaging (fMRI) and cardiovascular data were gathered from 34 healthy volunteers recruited within the human connectome project (the '100-unrelated subjects' release). Group-wise multi-level fMRI analyses in conjunction with complex instantaneous heartbeat correlates (entropy and Lyapunov exponent) revealed the existence of a specialized brain network, i.e. a complex central autonomic network (CCAN), reflecting what we refer to as complex autonomic control of the heart. Our results reveal CCAN areas comprised the paracingulate and cingulate gyri, temporal gyrus, frontal orbital cortex, planum temporale, temporal fusiform, superior and middle frontal gyri, lateral occipital cortex, angular gyrus, precuneous cortex, frontal pole, intracalcarine and supracalcarine cortices, parahippocampal gyrus and left hippocampus. The CCAN visible at rest does not include the insular cortex, thalamus, putamen, amygdala and right caudate, which are classical CAN regions peculiar to sympatho-vagal control. Our results also suggest that the CCAN is mainly involved in complex vagal control mechanisms, with possible links with emotional processing networks.