3D dissimilar-siamese-u-net for hyperdense Middle cerebral artery sign segmentation.

The hyperdense middle cerebral artery sign (HMCAS) representing a thromboembolus has been declared as a vital CT finding for intravascular thrombus in the diagnosis of acute ischemia stroke. Early recognition of HMCAS can assist in patient triage and subsequent thrombolysis or thrombectomy treatment. A total of 624 annotated head non-contrast-enhanced CT (NCCT) image scans were retrospectively collected from multiple public hospitals in Hong Kong. In this study, we present a deep Dissimilar-Siamese-U-Net (DSU-Net) that is able to precisely segment the lesions by integrating Siamese and U-Net architectures. The proposed framework consists of twin sub-networks that allow inputs of left and right hemispheres in head NCCT images separately. The proposed Dissimilar block fully explores the feature representation of the differences between the bilateral hemispheres. Ablation studies were carried out to validate the performance of various components of the proposed DSU-Net. Our findings reveal that the proposed DSU-Net provides a novel approach for HMCAS automatic segmentation and it outperforms the baseline U-Net and many state-of-the-art models for clinical practice.

Applying text mining methods to suicide research.

OBJECTIVE: To introduce the research methods of computerized text mining and its possible applications in suicide research and to demonstrate the procedures of applying a specific text mining area, document classification, to a suicide-related study. METHOD: A systematic search of academic papers that applied text mining methods to suicide research was conducted. Relevant papers were reviewed focusing on their research objectives and sources of data. Furthermore, a case of using natural language processing and document classification methods to analyze a large amount of suicide news was elaborated to showcase the methods. RESULTS: Eighty-six papers using text mining methods for suicide research have been published since 2001. The most common research objective (72.1%) was to classify which documents exhibit suicide risk or were written by suicidal people. The most frequently used data source was online social media posts (45.3%), followed by e-healthcare records (25.6%). For the news classification case, the top three classifiers trained for classification tasks achieved 84% or higher accuracy. CONCLUSIONS: Computerized text mining methods can help to scale up content analysis capacity and efficiency and uncover new insights and perspectives for suicide research.

Development of a data-driven COVID-19 prognostication tool to inform triage and step-down care for hospitalised patients in Hong Kong: a population-based cohort study.

BACKGROUND: This is the first study on prognostication in an entire cohort of laboratory-confirmed COVID-19 patients in the city of Hong Kong. Prognostic tool is essential in the contingency response for the next wave of outbreak. This study aims to develop prognostic models to predict COVID-19 patients' clinical outcome on day 1 and day 5 of hospital admission. METHODS: We did a retrospective analysis of a complete cohort of 1037 COVID-19 laboratory-confirmed patients in Hong Kong as of 30 April 2020, who were admitted to 16 public hospitals with their data sourced from an integrated electronic health records system. It covered demographic information, chronic disease(s) history, presenting symptoms as well as the worst clinical condition status, biomarkers' readings and Ct value of PCR tests on Day-1 and Day-5 of admission. The study subjects were randomly split into training and testing datasets in a 8:2 ratio. Extreme Gradient Boosting (XGBoost) model was used to classify the training data into three disease severity groups on Day-1 and Day-5. RESULTS: The 1037 patients had a mean age of 37.8 (SD ± 17.8), 53.8% of them were male. They were grouped under three disease outcome: 4.8% critical/serious, 46.8% stable and 48.4% satisfactory. Under the full models, 30 indicators on Day-1 and Day-5 were used to predict the patients' disease outcome and achieved an accuracy rate of 92.3% and 99.5%. With a trade-off between practical application and predictive accuracy, the full models were reduced into simpler models with seven common specific predictors, including the worst clinical condition status (4-level), age group, and five biomarkers, namely, CRP, LDH, platelet, neutrophil/lymphocyte ratio and albumin/globulin ratio. Day-1 model's accuracy rate, macro-/micro-averaged sensitivity and specificity were 91.3%, 84.9%/91.3% and 96.0%/95.7% respectively, as compared to 94.2%, 95.9%/94.2% and 97.8%/97.1% under Day-5 model. CONCLUSIONS: Both Day-1 and Day-5 models can accurately predict the disease severity. Relevant clinical management could be planned according to the predicted patients' outcome. The model is transformed into a simple online calculator to provide convenient clinical reference tools at the point of care, with an aim to inform clinical decision on triage and step-down care.

Automated Hierarchy Evaluation System of Large Vessel Occlusion in Acute Ischemia Stroke.

BACKGROUND: The detection of large vessel occlusion (LVO) plays a critical role in the diagnosis and treatment of acute ischemic stroke (AIS). Identifying LVO in the pre-hospital setting or early stage of hospitalization would increase the patients' chance of receiving appropriate reperfusion therapy and thereby improve neurological recovery. METHODS: To enable rapid identification of LVO, we established an automated evaluation system based on all recorded AIS patients in Hong Kong Hospital Authority's hospitals in 2016. The 300 study samples were randomly selected based on a disproportionate sampling plan within the integrated electronic health record system, and then separated into a group of 200 patients for model training, and another group of 100 patients for model performance evaluation. The evaluation system contained three hierarchical models based on patients' demographic data, clinical data and non-contrast CT (NCCT) scans. The first two levels of modeling utilized structured demographic and clinical data, while the third level involved additional NCCT imaging features obtained from deep learning model. All three levels' modeling adopted multiple machine learning techniques, including logistic regression, random forest, support vector machine (SVM), and eXtreme Gradient Boosting (XGboost). The optimal cut-off for the likelihood of LVO was determined by the maximal Youden index based on 10-fold cross-validation. Comparisons of performance on the testing group were made between these techniques. RESULTS: Among the 300 patients, there were 160 women and 140 men aged from 27 to 104 years (mean 76.0 with standard deviation 13.4). LVO was present in 130 (43.3%) patients. Together with clinical and imaging features, the XGBoost model at the third level of evaluation achieved the best model performance on testing group. The Youden index, accuracy, sensitivity, specificity, F1 score, and area under the curve (AUC) were 0.638, 0.800, 0.953, 0.684, 0.804, and 0.847, respectively. CONCLUSION: To the best of our knowledge, this is the first study combining both structured clinical data with non-structured NCCT imaging data for the diagnosis of LVO in the acute setting, with superior performance compared to previously reported approaches. Our system is capable of automatically providing preliminary evaluations at different pre-hospital stages for potential AIS patients.

Developing an early warning system of suicide using Google Trends and media reporting.

BACKGROUND: Conventional surveillance systems for suicides typically suffer from a substantial time lag of six months to two years. This study aims to develop an early warning system of possible suicide outbreaks in Hong Kong using Google Trends and suicide-related media reporting. METHODS: Data on 3,534 suicides from 2011 to 2015 were obtained from Hong Kong Census and Statistics Department, and the Coroner's Court. Using data from Google Trends and features extracted from media reporting on suicide news, we fitted Poisson regression models to predict the number and estimate the intensity of suicides on a weekly basis, for six subgroups, defined by gender and age. We adopted the cumulative sum (CUSUM) control chart-based method to identify outbreaks of suicide. RESULTS: The proposed model was able to predict the number of suicides with reasonably low normalized root mean squared errors, ranging from 15.6% for young females to 24.16% for old females. The suicide intensity curves were well captured by the proposed models for young males and females, but not for other groups. The Sensitivity, Precision and F1 Score of the CUSUM-based method were 50%, 100% and 67% for young females, and 93%, 54% and 68% for young males. LIMITATIONS: This study focused only on predicting the number of suicides in the current week, not in the future weeks. The model did not include social media, socioeconomic and climate data. CONCLUSIONS: Our results indicate that Google Trends search terms and media reporting data may be valuable data sources for predicting possible outbreak of suicides in Hong Kong. The proposed system could support effective and targeted interventions.

Female labour force participation and suicide rates in the world.

The current study aims to illustrate male to female suicide rate ratios in the world and explore the correlations between female labour force participation rates (FLPR) and suicide rates of both genders. Further, whether the relationship of FLPR and suicide rates vary according to the human capabilities of a given country are examined. Using suicide data obtained from the World Health Organization Statistical Information System, suicide gender ratios of 70 countries are illustrated. Based on the level of Human Development Index (HDI) and FLPR, the Bayesian Information Criteria (BIC) was used to determine the optimal number of clusters of those countries. Graphic illustrations of FLPR and gender-specific suicide rates, stratified by each cluster were presented, and Pearson's correlation coefficients calculated. Three clusters are identified, there was no correlation between FLPR and suicide rates in the first cluster where both the HDI and FLPR were the highest (Male: r = 0.29, P = 0.45; Female: r = 0.01, P = 0.97); whereas in Cluster 2, higher level of FLPR corresponded to lower suicide rates in both genders, although the statistical significance was only found in females (Male: r = -0.32, P = 0.15; Female: r = -0.48, P = 0.03). In Cluster 3 countries where HDI/FLPR were relatively lower, increased FLPR was associated with higher suicide rates for both genders (Male: r = 0.32, P = 0.04; Female: r = 0.32, P = 0.05). The relationship between egalitarian gender norms and suicide rates varies according to national context. A greater egalitarian gender norms may benefit both genders, but more so for women in countries equipped with better human capabilities. Although the beneficial effect may reach a plateau in countries with the highest HDI/FLPR; whereas in countries with relatively lower HDI/FLPR, increased FLPR were associated with higher suicide rates.