Latent Transformer Models for out-of-distribution detection.

Any clinically-deployed image-processing pipeline must be robust to the full range of inputs it may be presented with. One popular approach to this challenge is to develop predictive models that can provide a measure of their uncertainty. Another approach is to use generative modelling to quantify the likelihood of inputs. Inputs with a low enough likelihood are deemed to be out-of-distribution and are not presented to the downstream predictive model. In this work, we evaluate several approaches to segmentation with uncertainty for the task of segmenting bleeds in 3D CT of the head. We show that these models can fail catastrophically when operating in the far out-of-distribution domain, often providing predictions that are both highly confident and wrong. We propose to instead perform out-of-distribution detection using the Latent Transformer Model: a VQ-GAN is used to provide a highly compressed latent representation of the input volume, and a transformer is then used to estimate the likelihood of this compressed representation of the input. We demonstrate this approach can identify images that are both far- and near- out-of-distribution, as well as provide spatial maps that highlight the regions considered to be out-of-distribution. Furthermore, we find a strong relationship between an image's likelihood and the quality of a model's segmentation on it, demonstrating that this approach is viable for filtering out unsuitable images.

Antiepileptogenesis after stroke-trials and tribulations: Methodological challenges and recruitment results of a Phase II study with eslicarbazepine acetate.

There is currently no evidence to support the use of antiseizure medications to prevent unprovoked seizures following stroke. Experimental animal models suggested a potential antiepileptogenic effect for eslicarbazepine acetate (ESL), and a Phase II, multicenter, randomized, double-blind, placebo-controlled study was designed to test this hypothesis and assess whether ESL treatment for 1 month can prevent unprovoked seizures following stroke. We outline the design and status of this antiepileptogenesis study, and discuss the challenges encountered in its execution to date. Patients at high risk of developing unprovoked seizures after acute intracerebral hemorrhage or acute ischemic stroke were randomized to receive ESL 800 mg/d or placebo, initiated within 120 hours after primary stroke occurrence. Treatment continued until Day 30, then tapered off. Patients could receive all necessary therapies for stroke treatment according to clinical practice guidelines and standard of care, and are being followed up for 18 months. The primary efficacy endpoint is the occurrence of a first unprovoked seizure within 6 months after randomization ("failure rate"). Secondary efficacy assessments include the occurrence of a first unprovoked seizure during 12 months after randomization and during the entire study; functional outcomes (Barthel Index original 10-item version; National Institutes of Health Stroke Scale); post-stroke depression (Patient Health Questionnaire-9; PHQ-9); and overall survival. Safety assessments include the evaluation of treatment-emergent adverse events; laboratory parameters; vital signs; electrocardiogram; suicidal ideation and behavior (PHQ-9 question 9). The protocol aimed to randomize approximately 200 patients (1:1), recruited from 21 sites in seven European countries and Israel. Despite the challenges encountered, particularly during the COVID-19 pandemic, the study progressed and included a remarkable number of patients, with 129 screened and 125 randomized. Recruitment was stopped after 30 months, the first patient entered in May 2019, and the study is ongoing and following up on patients according to the Clinical Trial Protocol.

Dural arteriovenous fistula causing reversible cognitive impairment.

A previously independent 82-year-old woman presented with 5 months of worsening confusion, mobility and cognitive decline, with deficits in orientation, language and executive function. A cerebral dural arteriovenous fistula was identified and successfully embolised, after which her cognitive ability and independence dramatically improved. Although rare, a dural arteriovenous fistula may mimic a rapidly progressive dementia, but its early recognition and treatment can completely reverse the dementia.

Machine learning-enabled multitrust audit of stroke comorbidities using natural language processing.

BACKGROUND AND PURPOSE: With the increasing adoption of electronic records in the health system, machine learning-enabled techniques offer the opportunity for greater computer-assisted curation of these data for audit and research purposes. In this project, we evaluate the consistency of traditional curation methods used in routine clinical practice against a new machine learning-enabled tool, MedCAT, for the extraction of the stroke comorbidities recorded within the UK's Sentinel Stroke National Audit Programme (SSNAP) initiative. METHODS: A total of 2327 stroke admission episodes from three different National Health Service (NHS) hospitals, between January 2019 and April 2020, were included in this evaluation. In addition, current clinical curation methods (SSNAP) and the machine learning-enabled method (MedCAT) were compared against a subsample of 200 admission episodes manually reviewed by our study team. Performance metrics of sensitivity, specificity, precision, negative predictive value, and F1 scores are reported. RESULTS: The reporting of stroke comorbidities with current clinical curation methods is good for atrial fibrillation, hypertension, and diabetes mellitus, but poor for congestive cardiac failure. The machine learning-enabled method, MedCAT, achieved better performances across all four assessed comorbidities compared with current clinical methods, predominantly driven by higher sensitivity and F1 scores. CONCLUSIONS: We have shown machine learning-enabled data collection can support existing clinical and service initiatives, with the potential to improve the quality and speed of data extraction from existing clinical repositories. The scalability and flexibility of these new machine-learning tools, therefore, present an opportunity to revolutionize audit and research methods.

Rates, risks and routes to reduce vascular dementia (R4vad), a UK-wide multicentre prospective observational cohort study of cognition after stroke: Protocol.

BACKGROUND: Stroke commonly affects cognition and, by definition, much vascular dementia follows stroke. However, there are fundamental limitations in our understanding of vascular cognitive impairment, restricting understanding of prevalence, trajectories, mechanisms, prevention, treatment and patient-service needs. AIMS: Rates, Risks and Routes to Reduce Vascular Dementia (R4VaD) is an observational cohort study of post-stroke cognition. We aim to recruit a wide range of patients with stroke, presenting to geographically diverse UK hospitals, into a longitudinal study to determine rates of, and risk factors for, cognitive and related impairments after stroke, to assess potential mechanisms and improve prediction models. METHODS: We will recruit at least 2000 patients within six weeks of stroke with or without capacity to consent and collect baseline demographic, clinical, socioeconomic, lifestyle, cognitive, neuropsychiatric and informant data using streamlined patient-centred methods appropriate to the stage after stroke. We will obtain more detailed assessments at four to eight weeks after the baseline assessment and follow-up by phone and post yearly to at least two years. We will assess diagnostic neuroimaging in all and high-sensitivity inflammatory markers, genetics, blood pressure and diffusion tensor imaging in mechanistic sub-studies.Planned outputs: R4VaD will provide reliable data on long-term cognitive function after stroke, stratified by prior cognition, stroke- and patient-related variables and improved risk prediction. It will create a platform enabling sharing of data, imaging and samples. Participants will be consented for re-contact, facilitating future clinical trials and providing a resource for the stroke and dementia research communities.

A brain origin for factitious disorder (Munchausen's) with malingering? A single case with an old frontal lobe lesion.

We describe a patient who presented with gender identity dysphoria and stroke-like symptoms who we diagnosed with Munchausen's syndrome (factitious disorder). We discuss whether a brain lesion in the left frontal cortex is a possible risk factor, and support this hypothesis through neuropsychological investigation, EEG abnormalities, and a personality assessment. This case report supports previous suggestions that underlying brain disease/lesions might be risk factors for Munchausen's syndrome (factitious disorder).

Quantifying the Impact of Chronic Ischemic Injury on Clinical Outcomes in Acute Stroke With Machine Learning.

Acute stroke is often superimposed on chronic damage from previous cerebrovascular events. This background will inevitably modulate the impact of acute injury on clinical outcomes to an extent that will depend on the precise anatomical pattern of damage. Previous attempts to quantify such modulation have employed only reductive models that ignore anatomical detail. The combination of automated image processing, large-scale data, and machine learning now enables us to quantify the impact of this with high-dimensional multivariate models sensitive to individual variations in the detailed anatomical pattern. We introduce and validate a new automated chronic lesion segmentation routine for use with non-contrast CT brain scans, combining non-parametric outlier-detection score, Zeta, with an unsupervised 3-dimensional maximum-flow, minimum-cut algorithm. The routine was then applied to a dataset of 1,704 stroke patient scans, obtained at their presentation to a hyper-acute stroke unit (St George's Hospital, London, UK), and used to train a support vector machine (SVM) model to predict between low (0-2) and high (3-6) pre-admission and discharge modified Rankin Scale (mRS) scores, quantifying performance by the area under the receiver operating curve (AUROC). In this single center retrospective observational study, our SVM models were able to differentiate between low (0-2) and high (3-6) pre-admission and discharge mRS scores with an AUROC of 0.77 (95% confidence interval of 0.74-0.79), and 0.76 (0.74-0.78), respectively. The chronic lesion segmentation routine achieved a mean (standard deviation) sensitivity, specificity and Dice similarity coefficient of 0.746 (0.069), 0.999 (0.001), and 0.717 (0.091), respectively. We have demonstrated that machine learning models capable of capturing the high-dimensional features of chronic injuries are able to stratify patients-at the time of presentation-by pre-admission and discharge mRS scores. Our fully automated chronic stroke lesion segmentation routine simplifies this process, and utilizes routinely collected CT head scans, thereby facilitating future large-scale studies to develop supportive clinical decision tools.

A fast pathway for fear in human amygdala.

A fast, subcortical pathway to the amygdala is thought to have evolved to enable rapid detection of threat. This pathway's existence is fundamental for understanding nonconscious emotional responses, but has been challenged as a result of a lack of evidence for short-latency fear-related responses in primate amygdala, including humans. We recorded human intracranial electrophysiological data and found fast amygdala responses, beginning 74-ms post-stimulus onset, to fearful, but not neutral or happy, facial expressions. These responses had considerably shorter latency than fear responses that we observed in visual cortex. Notably, fast amygdala responses were limited to low spatial frequency components of fearful faces, as predicted by magnocellular inputs to amygdala. Furthermore, fast amygdala responses were not evoked by photographs of arousing scenes, which is indicative of selective early reactivity to socially relevant visual information conveyed by fearful faces. These data therefore support the existence of a phylogenetically old subcortical pathway providing fast, but coarse, threat-related signals to human amygdala.

Human brain lesion-deficit inference remapped.

Our knowledge of the anatomical organization of the human brain in health and disease draws heavily on the study of patients with focal brain lesions. Historically the first method of mapping brain function, it is still potentially the most powerful, establishing the necessity of any putative neural substrate for a given function or deficit. Great inferential power, however, carries a crucial vulnerability: without stronger alternatives any consistent error cannot be easily detected. A hitherto unexamined source of such error is the structure of the high-dimensional distribution of patterns of focal damage, especially in ischaemic injury-the commonest aetiology in lesion-deficit studies-where the anatomy is naturally shaped by the architecture of the vascular tree. This distribution is so complex that analysis of lesion data sets of conventional size cannot illuminate its structure, leaving us in the dark about the presence or absence of such error. To examine this crucial question we assembled the largest known set of focal brain lesions (n = 581), derived from unselected patients with acute ischaemic injury (mean age = 62.3 years, standard deviation = 17.8, male:female ratio = 0.547), visualized with diffusion-weighted magnetic resonance imaging, and processed with validated automated lesion segmentation routines. High-dimensional analysis of this data revealed a hidden bias within the multivariate patterns of damage that will consistently distort lesion-deficit maps, displacing inferred critical regions from their true locations, in a manner opaque to replication. Quantifying the size of this mislocalization demonstrates that past lesion-deficit relationships estimated with conventional inferential methodology are likely to be significantly displaced, by a magnitude dependent on the unknown underlying lesion-deficit relationship itself. Past studies therefore cannot be retrospectively corrected, except by new knowledge that would render them redundant. Positively, we show that novel machine learning techniques employing high-dimensional inference can nonetheless accurately converge on the true locus. We conclude that current inferences about human brain function and deficits based on lesion mapping must be re-evaluated with methodology that adequately captures the high-dimensional structure of lesion data.

A new method for automated high-dimensional lesion segmentation evaluated in vascular injury and applied to the human occipital lobe.

Making robust inferences about the functional neuroanatomy of the brain is critically dependent on experimental techniques that examine the consequences of focal loss of brain function. Unfortunately, the use of the most comprehensive such technique-lesion-function mapping-is complicated by the need for time-consuming and subjective manual delineation of the lesions, greatly limiting the practicability of the approach. Here we exploit a recently-described general measure of statistical anomaly, zeta, to devise a fully-automated, high-dimensional algorithm for identifying the parameters of lesions within a brain image given a reference set of normal brain images. We proceed to evaluate such an algorithm in the context of diffusion-weighted imaging of the commonest type of lesion used in neuroanatomical research: ischaemic damage. Summary performance metrics exceed those previously published for diffusion-weighted imaging and approach the current gold standard-manual segmentation-sufficiently closely for fully-automated lesion-mapping studies to become a possibility. We apply the new method to 435 unselected images of patients with ischaemic stroke to derive a probabilistic map of the pattern of damage in lesions involving the occipital lobe, demonstrating the variation of anatomical resolvability of occipital areas so as to guide future lesion-function studies of the region.

The effects of the dopamine agonist rotigotine on hemispatial neglect following stroke.

Hemispatial neglect following right-hemisphere stroke is a common and disabling disorder, for which there is currently no effective pharmacological treatment. Dopamine agonists have been shown to play a role in selective attention and working memory, two core cognitive components of neglect. Here, we investigated whether the dopamine agonist rotigotine would have a beneficial effect on hemispatial neglect in stroke patients. A double-blind, randomized, placebo-controlled ABA design was used, in which each patient was assessed for 20 testing sessions, in three phases: pretreatment (Phase A1), on transdermal rotigotine for 7-11 days (Phase B) and post-treatment (Phase A2), with the exact duration of each phase randomized within limits. Outcome measures included performance on cancellation (visual search), line bisection, visual working memory, selective attention and sustained attention tasks, as well as measures of motor control. Sixteen right-hemisphere stroke patients were recruited, all of whom completed the trial. Performance on the Mesulam shape cancellation task improved significantly while on rotigotine, with the number of targets found on the left side increasing by 12.8% (P = 0.012) on treatment and spatial bias reducing by 8.1% (P = 0.016). This improvement in visual search was associated with an enhancement in selective attention but not on our measures of working memory or sustained attention. The positive effect of rotigotine on visual search was not associated with the degree of preservation of prefrontal cortex and occurred even in patients with significant prefrontal involvement. Rotigotine was not associated with any significant improvement in motor performance. This proof-of-concept study suggests a beneficial role of dopaminergic modulation on visual search and selective attention in patients with hemispatial neglect following stroke.

Functional neuroanatomy: the locus of human intelligence.

A new study mapping the functional effects of brain lesions has revealed a surprising map of human intelligence, stimulating a re-evaluation of data from purely correlative methods such as functional magnetic resonance imaging.