Classification of cardioembolic stroke based on a deep neural network using chest radiographs.

BACKGROUND: Although chest radiographs have not been utilised well for classifying stroke subtypes, they could provide a plethora of information on cardioembolic stroke. This study aimed to develop a deep convolutional neural network that could diagnose cardioembolic stroke based on chest radiographs. METHODS: Overall, 4,064 chest radiographs of consecutive patients with acute ischaemic stroke were collected from a prospectively maintained stroke registry. Chest radiographs were randomly partitioned into training/validation (n = 3,255) and internal test (n = 809) datasets in an 8:2 ratio. A densely connected convolutional network (ASTRO-X) was trained to diagnose cardioembolic stroke based on chest radiographs. The performance of ASTRO-X was evaluated using the area under the receiver operating characteristic curve. Gradient-weighted class activation mapping was used to evaluate the region of focus of ASTRO-X. External testing was performed with 750 chest radiographs of patients with acute ischaemic stroke from 7 hospitals. FINDINGS: The areas under the receiver operating characteristic curve of ASTRO-X were 0.86 (95% confidence interval [CI], 0.83-0.89) and 0.82 (95% CI, 0.79-0.85) during the internal and multicentre external testing, respectively. The gradient-weighted class activation map demonstrated that ASTRO-X was focused on the area where the left atrium was located. Compared with cases predicted as non-cardioembolism by ASTRO-X, cases predicted as cardioembolism by ASTRO-X had higher left atrial volume index and lower left ventricular ejection fraction in echocardiography. INTERPRETATION: ASTRO-X, a deep neural network developed to diagnose cardioembolic stroke based on chest radiographs, demonstrated good classification performance and biological plausibility. FUNDING: Grant No. 14-2020-046 and 08-2016-051 from the Seoul National University Bundang Research Fund and NRF-2020M3E5D9079768 from the National Research Foundation of Korea.

[Embolic stroke of undetermined source (ESUS) - Classification of a new stroke entity]. / Embolischer Schlaganfall mit ungeklärter Emboliequelle (ESUS) ­ Klassifikation einer neuen Schlaganfallentität.

Embolic stroke of undetermined source (ESUS) represents a subpopulation of cryptogenic strokes defined by its embolic stroke pattern on imaging and if after a carefully performed diagnostic evaluation, a specific, well recognized cause of stroke has not been identified. This review article analyses the basics of the ESUS concept and provides an overview of the evidence from recent cohort studies. The definition, aetiology and diagnosis of ESUS are reassessed. Targeted diagnostics in ESUS patients can reduce the number of cryptogenic strokes by making a specific diagnosis.

Inconsistent centralised versus non-centralised ischaemic stroke aetiology.

BACKGROUND AND PURPOSE: The Trial of Org 10 172 in Acute Stroke Treatment (TOAST) system is the most widely used aetiological categorisation system in clinical practice and research. Limited studies have validated the accuracy of routine aetiological diagnosis of patients with ischaemic stroke according to the TOAST criteria when the reported subtype is assumed to be correct. We investigated the agreement between centralised and non-centralised (site-reported, at discharge) stroke subtypes in the Third China National Stroke Registry (CNSR-III), and analysed the influence of classification consistency on evaluation during hospitalisation and for secondary prevention strategy. METHODS: All patients with ischaemic stroke from the CNSR-III study with complete diffusion-weighted imaging data were included. We used multivariable Cox proportional-hazard regression models to evaluate the factors associated with consistency between centralised and non-centralised stroke subtypes. Sensitivity analyses were conducted of the subgroup of patients with complete information. RESULTS: This study included 12 180 patients (mean age, 62.3 years; and women, 31.4%). Agreement between centralised and non-centralised subtype was the highest for the large-artery atherosclerosis subtype stroke (77.4% of centralised patients), followed by the small-vessel occlusion subtype (40.6% of centralised patients). Agreements for cardioembolism and stroke of other determined aetiology subtypes were 38.7% and 12.2%, respectively. Patient-level and hospital-level factors were associated with the inconsistency between centralised/non-centralised aetiological subtyping. This inconsistency was related to differences in secondary prevention strategies. Only 15.3% of the newly diagnosed patients with cardioembolism underwent centralised subtyping with indications to receive oral anticoagulants at discharge. In comparison, 51.3% of the consistent cardioembolism group and 42.0% of the centrally reassigned cardioembolism group with anticoagulation indications were prescribed oral anticoagulants. CONCLUSIONS: Substantial inconsistency exists between centralised and non-centralised subtyping in China. Inaccurate aetiological subtyping could lead to inadequate secondary prevention, especially in patients with cardioembolic stroke.

Phenotypic ASCOD characterisations of ischaemic stroke in the young at an urban tertiary care centre.

BACKGROUND AND PURPOSE: Stroke in young individuals is a serious public health burden. This study aimed to characterise the various phenotypes of ischaemic stroke in a young urban population (≤50 years old) using the ASCOD classification system, which assigns a score to five stroke categories: atherosclerosis, small vessel disease (SVD), cardioembolism, other and dissection. Within each category, a numerical score represents the degree of causality attributed to the stroke. METHODS: This retrospective study cohort was composed of patients from an urban tertiary care academic centre. Cases were selected by searching Get With the Guidelines database for adults ≤50 years old with ischaemic stroke. The study sample included 175 ischaemic strokes in 157 patients, with 16 subjects re-infarcting. Using retrospective chart review, each stroke was scored according to the ASCOD classification system. Multivariable logistic regression analyses were performed to explore each ASCOD category's association with causal risk factors. RESULTS: Of possible causal mechanisms, defined as receiving a grade 1 or 2, a cardiovascular aetiology was most prevalent (25.7%), followed by SVD (22.3%), and closely by atherosclerosis (21.1%). Of general phenotypes, defined as receiving a grade 1 or 2 or 3, atherosclerosis was the most prevalent (51.4%), followed by SVD (47.4%), cardioembolism (42.3%) and other (35.4%). 31.6% of all strokes were of unclear aetiology. Subjects between 45 and 50 years old were more likely to develop a cardioembolic or SVD stroke when compared with subjects <45 years old. CONCLUSION: This study took a novel approach to ASCOD phenotyping, allowing several observations: (1) In patients with advanced atherosclerosis receiving the score A1, the vast majority had systemic atherosclerosis in multiple vascular territories; (2) the cardiac score C2(6), defined as a radiographic pattern highly suggestive of a central embolic source, may overestimate the prevalence of true cardiac disease; (3) incidental laboratory findings may detect some underlying pathology, but causality to the stroke is unlikely.