Transformer for Sub-Seasonal Extreme High Temperature Probabilistic Forecasting Over Eastern China

Sub-seasonal high temperature forecasting is significant for early warning of extreme heat weather. Currently, deep learning methods, especially Transformer, have been successfully applied to the meteorological field. Relying on the excellent global feature extraction capability in natural language processing, Transformer may be useful to improve the ability in extended periods. To explore this, we introduce the Transformer and propose a Transformer-based model, called Transformer to High Temperature (T2T). In the details of the model, we successively discuss the use of the Transformer and the position encoding in T2T to continuously optimize the model structure in an experimental manner. In the dataset, the multi-version data fusion method is proposed to further improve the prediction of the model with reasonable expansion of the dataset. The performance of well-desinged model (T2T) is verified against the European Centre for Medium-Range Weather Forecasts (ECMWF) and Multi-Layer Perceptron (MLP) at each grid of the 100.5°E to 138°E, 21°N to 54°N domain for the April to October of 2016-2019. For case study initiated from 2 June 2018, the results indicated that T2T is significantly better than ECMWF and MLP, with smaller absolute error and more reliable probabilistic forecast for the extreme high event happened during the third week. Over all, the deterministic forecast of T2T is superior to MLP and ECMWF due to ability of utilize spatial information of grids. T2T also provided a better calibrated probability of high temperature and a sharper prediction probability density function than MLP and ECMWF. All in all, T2T can meet the operational requirements for extended period forecasting of extreme high temperature. Furthermore, our research can provide experience on the development of deep learning in this field and achieve the continuous progress of seamless forecasting systems.

The Clinical Value of Explainable Deep Learning for Diagnosing Fungal Keratitis Using in vivo Confocal Microscopy Images

Background: Artificial intelligence (AI) has great potential to detect fungal keratitis using in vivo confocal microscopy images, but its clinical value remains unclarified. A major limitation of its clinical utility is the lack of explainability and interpretability.Methods: An explainable AI (XAI) system based on Gradient-weighted Class Activation Mapping (Grad-CAM) and Guided Grad-CAM was established. In this randomized controlled trial, nine ophthalmologists (three expert ophthalmologists, three competent ophthalmologists, and three novice ophthalmologists) read images in each of the conditions: unassisted, AI-assisted, or XAI-assisted. In unassisted condition, only the original IVCM images were shown to the readers. AI assistance comprised a histogram of model prediction probability. For XAI assistance, explanatory maps were additionally shown. The accuracy, sensitivity, and specificity were calculated against an adjudicated reference standard. Moreover, the time spent was measured.Results: Both forms of algorithmic assistance increased the accuracy and sensitivity of competent and novice ophthalmologists significantly without reducing specificity. The improvement was more pronounced in XAI-assisted condition than that in AI-assisted condition. Time spent with XAI assistance was not significantly different from that without assistance.Conclusion: AI has shown great promise in improving the accuracy of ophthalmologists. The inexperienced readers are more likely to benefit from the XAI system. With better interpretability and explainability, XAI-assistance can boost ophthalmologist performance beyond what is achievable by the reader alone or with black-box AI assistance.

Visualizing Phytochemical-Protein Interaction Networks: Momordica charantia and Cancer

The in silico study of medicinal plants is a rapidly growing field. Techniques such as reverse screening and network pharmacology are used to study the complex cellular action of medicinal plants against disease. However, it is difficult to produce a meaningful visualization of phytochemical-protein interactions (PCPIs) in the cell. This study introduces a novel workflow combining various tools to visualize a PCPI network for a medicinal plant against a disease. The five steps are 1) phytochemical compilation, 2) reverse screening, 3) network building, 4) network visualization, and 5) evaluation. The output is a PCPI network that encodes multiple dimensions of information, including subcellular location, phytochemical class, pharmacokinetic data, and prediction probability. As a proof of concept, we built a PCPI network for bitter gourd (Momordica charantia L.) against colorectal cancer. The network and workflow are available at https://yumibriones.github.io/network/. The PCPI network highlights high-confidence interactions for further in vitro or in vivo study. The overall workflow is broadly transferable and can be used to visualize the action of other medicinal plants or small molecules against other diseases.

Cohort study on the prognosis of acute cerebral infarction in different circulatory systems at 1-year follow-up

Background To evaluate the prognosis of acute cerebral infarction at 1-year follow-up in different circulation infarctions. Methods Clinical data of 858 consecutive patients with acute cerebral infarction were collected. Of the 858 cases, 21 (2.45%) were lost to follow-up and 837 completed follow-up and thus were enrolled in this study. At 1-year follow-up, death or moderate-to-severe dysfunction (modified Rankin Scale (mRS) ≥ 3 points) was regarded as the poor prognostic endpoint. Univariate analysis and multivariate logistic stepwise regression analysis were performed to assess the prognosis. The prediction probability of indicators was obtained for the multivariate model, and the receiver operating characteristic curve was delineated to calculate the area under the curve (AUC) to predict the fitness of the model. Results The older the age, the greater the probability of a poor prognosis. Patients with previous diabetes and cerebral infarction had a poor prognosis. The higher the National Institutes of Health Stroke Scale and mRS scores and the lower the Barthel index at admission, the worse the prognosis of the patients. The longer the hospital stay, the worse the prognosis of the patients. The prognosis of different circulation infarctions was different. The AUC of the multivariate model was AUC = 0.893, and the 95% confidence interval was 0.870–0.913, indicating a good fit. The prognosis of anterior circulation infarction (ACI) was worse than that of posterior circulation infarction (PCI) (P < 0.05). The prognosis of patients with ACI and PCI was not significantly different from that of patients with ACI or PCI alone (P > 0.05). Conclusions Diabetes, the Barthel index at admission and previous cerebral infarction are poor prognostic factors of acute cerebral infarction. The prognosis of ACI is worse than that of PCI. Different factors affect the prognosis of different circulatory system infarctions.

Radiomics-based prediction of multiple gene alteration incorporating mutual genetic information in glioblastoma and grade 4 astrocytoma, IDH-mutant

Purpose In glioma, molecular alterations are closely associated with disease prognosis. This study aimed to develop a radiomics-based multiple gene prediction model incorporating mutual information of each genetic alteration in glioblastoma and grade 4 astrocytoma, IDH-mutant. Methods From December 2014 through January 2020, we enrolled 418 patients with pathologically confirmed glioblastoma (based on the 2016 WHO classification). All selected patients had preoperative MRI and isocitrate dehydrogenase (IDH) mutation, O-6-methylguanine-DNA methyltransferase (MGMT) promoter methylation, epidermal growth factor receptor amplification, and alpha-thalassemia/mental retardation syndrome X-linked (ATRX) loss status. Patients were randomly split into training and test sets (7:3 ratio). Enhancing tumor and peritumoral T2-hyperintensity were auto-segmented, and 660 radiomics features were extracted. We built binary relevance (BR) and ensemble classifier chain (ECC) models for multi-label classification and compared their performance. In the classifier chain, we calculated the mean absolute Shapley value of input features. Results The micro-averaged area under the curves (AUCs) for the test set were 0.804 and 0.842 in BR and ECC models, respectively. IDH mutation status was predicted with the highest AUCs of 0.964 (BR) and 0.967 (ECC). The ECC model showed higher AUCs than the BR model for ATRX (0.822 vs. 0.775) and MGMT promoter methylation (0.761 vs. 0.653) predictions. The mean absolute Shapley values suggested that predicted outcomes from the prior classifiers were important for better subsequent predictions along the classifier chains. Conclusion We built a radiomics-based multiple gene prediction chained model that incorporates mutual information of each genetic alteration in glioblastoma and grade 4 astrocytoma, IDH-mutant and performs better than a simple bundle of binary classifiers using prior classifiers’ prediction probability.

Verification of the performance of the Bispectral Index as a hypnotic depth indicator during dexmedetomidine sedation

Background: Differences in the effects of propofol and dexmedetomidine sedation on electroencephalogram patterns have been reported previously. However, the reliability of the Bispectral Index (BIS) value for assessing the sedation caused by dexmedetomidine remains debatable. The purpose of this study is to evaluate the correlation between the BIS value and the Modified Observer’s Assessment of Alertness/Sedation (MOAA/S) scale in patients sedated with dexmedetomidine. Methods: Forty-two patients (age range, 20–80 years) who were scheduled for elective surgery under spinal anesthesia were enrolled in this study. Spinal anesthesia was performed using 0.5% bupivacaine, which was followed by dexmedetomidine infusion (loading dose, 0.5–1 μg/kg for 10 min; maintenance dose, 0.3–0.6 μg/kg/h). The MOAA/S score was used to evaluate the level of sedation, and the Vital Recorder program was used to collect data (vital signs and BIS values). Results: A total of 215082 MOAA/S scores and BIS data pairs were analyzed. The baseline variability of the BIS value was 7.024%, and the decrease in the BIS value was associated with a decrease in the MOAA/S score. The correlation coefficient and prediction probability between the two measurements were 0.566 (P < 0.0001) and 0.636, respectively. The mean ± standard deviation values of the BIS were 87.22 ± 7.06, 75.85 ± 9.81, and 68.29 ± 12.65 when the MOAA/S scores were 5, 3, and 1, respectively. Furthermore, the cut-off BIS values in the receiver operating characteristic analysis at MOAA/S scores of 5, 3, and 1 were 82, 79, and 73, respectively. Conclusion: The BIS values were significantly correlated with the MOAA/S scores. Thus, the BIS along with the clinical sedation scale might prove useful in assessing the hypnotic depth of a patient during sedation with dexmedetomidine.

Performance of a 3D convolutional neural network in the detection of hypoperfusion at CT pulmonary angiography in patients with chronic pulmonary embolism: a feasibility study

Background Chronic pulmonary embolism (CPE) is a life-threatening disease easily misdiagnosed on computed tomography. We investigated a three-dimensional convolutional neural network (CNN) algorithm for detecting hypoperfusion in CPE from computed tomography pulmonary angiography (CTPA). Methods Preoperative CTPA of 25 patients with CPE and 25 without pulmonary embolism were selected. We applied a 48%–12%–40% training-validation-testing split (12 positive and 12 negative CTPA volumes for training, 3 positives and 3 negatives for validation, 10 positives and 10 negatives for testing). The median number of axial images per CTPA was 335 (min–max, 111–570). Expert manual segmentations were used as training and testing targets. The CNN output was compared to a method in which a Hounsfield unit (HU) threshold was used to detect hypoperfusion. Receiver operating characteristic area under the curve (AUC) and Matthew correlation coefficient (MCC) were calculated with their 95% confidence interval (CI). Results The predicted segmentations of CNN showed AUC 0.87 (95% CI 0.82–0.91), those of HU-threshold method 0.79 (95% CI 0.74–0.84). The optimal global threshold values were CNN output probability ≥ 0.37 and ≤ -850 HU. Using these values, MCC was 0.46 (95% CI 0.29–0.59) for CNN and 0.35 (95% CI 0.18–0.48) for HU-threshold method (average difference in MCC in the bootstrap samples 0.11 (95% CI 0.05–0.16). A high CNN prediction probability was a strong predictor of CPE. Conclusions We proposed a deep learning method for detecting hypoperfusion in CPE from CTPA. This model may help evaluating disease extent and supporting treatment planning.

Ensemble model for Heart Disease Prediction

For the identification and prediction of different diseases, machine learning techniques are commonly used in clinical decision support systems. Since heart disease is the leading cause of death for both men and women around the world. Heart is one of the essential parts of human body, therefore, it is one of the most critical concerns in the medical domain, and several researchers have developed intelligent medical devices to support the systems and further to enhance the ability to diagnose and predict heart diseases. However, there are few studies that look at the capabilities of ensemble methods in developing a heart disease detection and prediction model. In this study, the researchers assessed that how to use ensemble model, which proposes a more stable performance than the use of base learning algorithm and these leads to better results than other heart disease prediction models. The University of California, Irvine (UCI) Machine Learning Repository archive was used to extract patient heart disease data records. To achieve the aim of this study, the researcher developed the meta-algorithm. The ensemble model is a superior solution in terms of high predictive accuracy and diagnostics output reliability, as per the results of the experiments. An ensemble heart disease prediction model is also presented in this work as a valuable, cost-effective, and timely predictive option with a user-friendly graphical user interface that is scalable and expandable. From the finding, the researcher suggests that Bagging is the best ensemble classifier to be adopted as the extended algorithm that has the high prediction probability score in the implementation of heart disease prediction.

A hybrid model with novel feature selection method and enhanced voting method for credit scoring

With the in-depth application of artificial intelligence technology in the financial field, credit scoring models constructed by machine learning algorithms have become mainstream. However, the high-dimensional and complex attribute features of the borrower pose challenges to the predictive competence of the model. This paper proposes a hybrid model with a novel feature selection method and an enhanced voting method for credit scoring. First, a novel feature selection combined method based on a genetic algorithm (FSCM-GA) is proposed, in which different classifiers are used to select features in combination with a genetic algorithm and combine them to generate an optimal feature subset. Furthermore, an enhanced voting method (EVM) is proposed to integrate classifiers, with the aim of improving the classification results in which the prediction probability values are close to the threshold. Finally, the predictive competence of the proposed model was validated on three public datasets and five evaluation metrics (accuracy, AUC, F-score, Log loss and Brier score). The comparative experiment and significance test results confirmed the good performance and robustness of the proposed model.

A Multichannel Deep Neural Network for Retina Vessel Segmentation via a Fusion Mechanism

Changes in fundus blood vessels reflect the occurrence of eye diseases, and from this, we can explore other physical diseases that cause fundus lesions, such as diabetes and hypertension complication. However, the existing computational methods lack high efficiency and precision segmentation for the vascular ends and thin retina vessels. It is important to construct a reliable and quantitative automatic diagnostic method for improving the diagnosis efficiency. In this study, we propose a multichannel deep neural network for retina vessel segmentation. First, we apply U-net on original and thin (or thick) vessels for multi-objective optimization for purposively training thick and thin vessels. Then, we design a specific fusion mechanism for combining three kinds of prediction probability maps into a final binary segmentation map. Experiments show that our method can effectively improve the segmentation performances of thin blood vessels and vascular ends. It outperforms many current excellent vessel segmentation methods on three public datasets. In particular, it is pretty impressive that we achieve the best F1-score of 0.8247 on the DRIVE dataset and 0.8239 on the STARE dataset. The findings of this study have the potential for the application in an automated retinal image analysis, and it may provide a new, general, and high-performance computing framework for image segmentation.