Ambient temperature and the occurrence of intradialytic hypotension in patients receiving hemodialysis.

Background: Intradialytic hypotension (IDH) is a common hemodialysis complication causing adverse outcomes. Despite the well-documented associations of ambient temperatures with fluid removal and pre-dialysis blood pressure (BP), the relationship between ambient temperature and IDH has not been adequately studied. Methods: We conducted a cohort study at a tertiary hospital in southern Taiwan between 1 January 2016 and 31 October 2021. The 24-h pre-hemodialysis mean ambient temperature was determined using hourly readings from the weather station closest to each patient's residence. IDH was defined using Fall40 [systolic BP (SBP) drop of ≥40 mmHg] or Nadir90/100 (SBP <100 if pre-dialysis SBP was ≥160, or SBP <90 mmHg). Multivariate logistic regression with generalizing estimating equations and mediation analysis were utilized. Results: The study examined 110 400 hemodialysis sessions from 182 patients, finding an IDH prevalence of 11.8% and 10.4% as per the Fall40 and Nadir90/100 criteria, respectively. It revealed a reverse J-shaped relationship between ambient temperature and IDH, with a turning point around 27°C. For temperatures under 27°C, a 4°C drop significantly increased the odds ratio of IDH to 1.292 [95% confidence interval (CI) 1.228 to 1.358] and 1.207 (95% CI 1.149 to 1.268) under the Fall40 and Nadir90/100 definitions, respectively. Lower ambient temperatures correlated with higher ultrafiltration, accounting for about 23% of the increased IDH risk. Stratified seasonal analysis indicated that this relationship was consistent in spring, autumn and winter. Conclusion: Lower ambient temperature is significantly associated with an increased risk of IDH below the threshold of 27°C, irrespective of the IDH definition. This study provides further insight into environmental risk factors for IDH in patients undergoing hemodialysis.

Development and validation of a machine learning model of radiation-induced hypothyroidism with clinical and dose-volume features.

BACKGROUND AND PURPOSE: Radiation-induced hypothyroidism (RIHT) is a common but underestimated late effect in head and neck cancers. However, no consensus exists regarding risk prediction or dose constraints in RIHT. We aimed to develop a machine learning model for the accurate risk prediction of RIHT based on clinical and dose-volume features and to evaluate its performance internally and externally. MATERIALS AND METHODS: We retrospectively searched two institutions for patients aged >20 years treated with definitive radiotherapy for nasopharyngeal or oropharyngeal cancer, and extracted their clinical information and dose-volume features. One was designated the developmental cohort, the other as the external validation cohort. We compared the performances of machine learning models with those of published normal tissue complication probability (NTCP) models. RESULTS: The developmental and external validation cohorts consisted of 378 and 49 patients, respectively. The estimated cumulative incidence rates of grade ≥1 hypothyroidism were 53.5% and 61.3% in the developmental and external validation cohorts, respectively. Machine learning models outperformed traditional NTCP models by having lower Brier scores at every time point and a lower integrated Brier score, while demonstrating a comparable calibration index and mean area under the curve. Even simplified machine learning models using only thyroid features performed better than did traditional NTCP algorithms. The machine learning models showed consistent performance between folds. The performance in a previously unseen external validation cohort was comparable to that of the cross-validation. CONCLUSIONS: Our model outperformed traditional NTCP models, with additional capabilities of predicting the RIHT risk at individual time points. A simplified model using only thyroid dose-volume features still outperforms traditional NTCP models and can be incorporated into future treatment planning systems for biological optimization.

Predicting hyperkalemia in patients with advanced chronic kidney disease using the XGBoost model.

BACKGROUND: Hyperkalemia is a common complication of chronic kidney disease (CKD). Hyperkalemia is associated with mortality, CKD progression, hospitalization, and high healthcare costs in patients with CKD. We developed a machine learning model to predict hyperkalemia in patients with advanced CKD at an outpatient clinic. METHODS: This retrospective study included 1,965 advanced CKD patients between January 1, 2010, and December 31, 2020 in Taiwan. We randomly divided all patients into the training (75%) and testing (25%) datasets. The primary outcome was to predict hyperkalemia (K+ > 5.5 mEq/L) in the next clinic vist. Two nephrologists were enrolled in a human-machine competition. The area under the receiver operating characteristic curves (AUCs), sensitivity, specificity, and accuracy were used to evaluate the performance of XGBoost and conventional logistic regression models with that of these physicians. RESULTS: In a human-machine competition of hyperkalemia prediction, the AUC, PPV, and accuracy of the XGBoost model were 0.867 (95% confidence interval: 0.840-0.894), 0.700, and 0.933, which was significantly better than that of our clinicians. There were four variables that were chosen as high-ranking variables in XGBoost and logistic regression models, including hemoglobin, the serum potassium level in the previous visit, angiotensin receptor blocker use, and calcium polystyrene sulfonate use. CONCLUSIONS: The XGBoost model provided better predictive performance for hyperkalemia than physicians at the outpatient clinic.

Histopathology images predict multi-omics aberrations and prognoses in colorectal cancer patients.

Histopathologic assessment is indispensable for diagnosing colorectal cancer (CRC). However, manual evaluation of the diseased tissues under the microscope cannot reliably inform patient prognosis or genomic variations crucial for treatment selections. To address these challenges, we develop the Multi-omics Multi-cohort Assessment (MOMA) platform, an explainable machine learning approach, to systematically identify and interpret the relationship between patients' histologic patterns, multi-omics, and clinical profiles in three large patient cohorts (n = 1888). MOMA successfully predicts the overall survival, disease-free survival (log-rank test P-value<0.05), and copy number alterations of CRC patients. In addition, our approaches identify interpretable pathology patterns predictive of gene expression profiles, microsatellite instability status, and clinically actionable genetic alterations. We show that MOMA models are generalizable to multiple patient populations with different demographic compositions and pathology images collected from distinctive digitization methods. Our machine learning approaches provide clinically actionable predictions that could inform treatments for colorectal cancer patients.

Literature-based translation from synthetic lethality screening into therapeutics targets: CD82 is a novel target for KRAS mutation in colon cancer.

Synthetic lethality (SL) is an emerging therapeutic paradigm in cancer. We introduced a different approach to prioritize SL gene pairs through literature mining and RAS-mutant high-throughput screening (HTS) data. We matched essential genes from text-mining and mutant genes from the COSMIC and CCLE HTS datasets to build a prediction model of SL gene pairs. CCLE gene expression data were used to enrich the essential-mutant SL gene pairs using Spearman's correlation coefficient and literature mining. In total, 223 essential trigger terms were extracted and ranked. The threshold of the essential gene score ( S g ) was set to 10. We identified 586 genes essential for the SL prediction model of colon cancer. Seven essential RAS-mutant SL gene pairs were identified in our model, including CD82-KRAS/NRAS, PEBP1-NRAS, MT-CO2-HRAS, IFI27-NRAS/KRAS, and SUMO1-HRAS gene pairs. Using RAS-mutant HTS data validation, we identified two potential SL gene pairs, including the CD82 (essential gene)-KRAS (mutant gene) pair and CD82-NRAS pair in the DLD-1 colon cancer cell line (Spearman's correlation p-values = 0.004786 and 0.00249, respectively). Based on further annotations by PubChem, we observed that digitonin targeted the complex comprising CD82, especially in KRAS-mutated HCT116 cancer cells. Moreover, we experimentally demonstrated that CD82 exhibited selective vulnerability in KRAS-mutant colorectal cancer. We used literature mining and HTS data to identify candidates for SL targets for RAS-mutant colon cancer.

Predicting Mortality Using Machine Learning Algorithms in Patients Who Require Renal Replacement Therapy in the Critical Care Unit.

Background: General severity of illness scores are not well calibrated to predict mortality among patients receiving renal replacement therapy (RRT) for acute kidney injury (AKI). We developed machine learning models to make mortality prediction and compared their performance to that of the Sequential Organ Failure Assessment (SOFA) and HEpatic failure, LactatE, NorepInephrine, medical Condition, and Creatinine (HELENICC) scores. Methods: We extracted routinely collected clinical data for AKI patients requiring RRT in the MIMIC and eICU databases. The development models were trained in 80% of the pooled dataset and tested in the rest of the pooled dataset. We compared the area under the receiver operating characteristic curves (AUCs) of four machine learning models (multilayer perceptron [MLP], logistic regression, XGBoost, and random forest [RF]) to that of the SOFA, nonrenal SOFA, and HELENICC scores and assessed calibration, sensitivity, specificity, positive (PPV) and negative (NPV) predicted values, and accuracy. Results: The mortality AUC of machine learning models was highest for XGBoost (0.823; 95% confidence interval [CI], 0.791−0.854) in the testing dataset, and it had the highest accuracy (0.758). The XGBoost model showed no evidence of lack of fit with the Hosmer−Lemeshow test (p > 0.05). Conclusion: XGBoost provided the highest performance of mortality prediction for patients with AKI requiring RRT compared with previous scoring systems.

Robot-enhanced diabetes care for middle-aged and older adults living with diabetes in the community: A small sample size mixed-method evaluation.

PURPOSE: This study assessed robot-enhanced healthcare in practical settings for the purpose of community diabetes care. METHODS: A mixed method evaluation collected quantitative and qualitative data on diabetes patients over 45 (N = 30) and community pharmacists (N = 10). It took 15-20 min for the diabetes patients to interact with the robot. Before and after the interaction, questionnaires including a diabetes knowledge test, self-efficacy for diabetes, and feasibility of use of the robot was administered. In-depth interviews with both pharmacists and patients were also conducted. RESULTS: After interacting with the robot, a statistically significant improvement in diabetes knowledge (p < .001) and feasibility of the robot (p = .012) was found, but self-efficacy (p = .171) was not significantly improved. Five themes emerged from interviewing the diabetes patients: Theme 1: meets the needs of self-directed learning for the elderly; Theme 2: reduces alertness and creates comfortable interaction; Theme 3: vividness and richness enhance interaction opportunities; Theme 4: Robots are not without disadvantages, and Theme 5: Every person has unique tastes. Three themes emerged from interviewing pharmacists: Theme 1: Technology must meet the real needs of the patient; Theme 2: creates new services, and Theme 3: The use of robots must conform to real-life situations. CONCLUSIONS: Both the diabetes patients and the pharmacist reported more positive feedback on the robot-enhanced diabetes care than concerns. Self-directed learning, comfortable interaction, and vividness were the most focuses when using robot to enhance self-management for the patients. Pharmacists were most receptive to fit conforming with reality and creating new services.

Possibilistic classification by support vector networks.

In many real-world classification problems, the available information is often uncertain. In order to effectively describe the inherent vagueness and improve the classification performance, this paper proposes a novel possibilistic classification algorithm using support vector machines (SVMs). Based on possibility theory, the proposed algorithm aims at finding a maximal-margin fuzzy hyperplane by solving a fuzzy mathematical optimization problem Moreover, the decision function of the proposed approach is generalized such that the values assigned to the data vectors fall within a specified range and indicate the membership grade of these data vectors in the positive class. The proposed algorithm retains the advantages of fuzzy set theory and SVM theory. The proposed approach is more robust for handling data corrupted by outliers. Moreover, the structural risk minimization principle of SVMs enables the proposed approach to effectively classify the unseen data. Furthermore, the proposed algorithm has additional advantage of using vagueness parameter v for controlling the bounds on fractions of support vectors and errors. The extensive experiments performed on benchmark datasets and real applications demonstrate that the proposed algorithm has satisfactory generalization accuracy and better describes the inherent vagueness in the given dataset.

DiaDeL: An Accurate Deep Learning-Based Model With Mutational Signatures for Predicting Metastasis Stage and Cancer Types.

Mutational signatures help identify cancer-associated genes that are being involved in tumorigenesis pathways. Hence, these pathways guide precision medicine approaches to find appropriate drugs and treatments. The pattern of mutations varies in different cancer types. Some mutations dysregulate protein function so that their accumulation is responsible for cancer development and might be associated with different cancer types. Therefore, mutations as a feature set can be used as an informative candidate to distinguish various cancer types. There are several options for demonstrating mutations. One might employ binary values to demonstrate mutation regions. Another potential method for extracting features is utilizing mutation interpreters. In this study, we investigate the trinucleotide mutational pattern of each cancer type. Moreover, we extract salient NMF-based mutational signatures across various cancer types. Then, we identify cancer-associated genes of a target cancer based on its salient signatures. We evaluate the cancer-associated genes using survival and gene expression analysis in different stages of cancer. Furthermore, we introduce DiaDeL, which is a deep learning-based binary classifier. The DiaDeL model uses mutational signatures as input features and distinct a cancer type from the others. Our proposed model outperforms six state-of-the-art methods with 0.824 and 0.88 for accuracy and AUC, respectively. The source code is available at https://github.com/sabdollahi/DiaDeL.

Contextualizing Genes by Using Text-Mined Co-Occurrence Features for Cancer Gene Panel Discovery.

Developing a biomedical-explainable and validatable text mining pipeline can help in cancer gene panel discovery. We create a pipeline that can contextualize genes by using text-mined co-occurrence features. We apply Biomedical Natural Language Processing (BioNLP) techniques for literature mining in the cancer gene panel. A literature-derived 4,679 × 4,630 gene term-feature matrix was built. The EGFR L858R and T790M, and BRAF V600E genetic variants are important mutation term features in text mining and are frequently mutated in cancer. We validate the cancer gene panel by the mutational landscape of different cancer types. The cosine similarity of gene frequency between text mining and a statistical result from clinical sequencing data is 80.8%. In different machine learning models, the best accuracy for the prediction of two different gene panels, including MSK-IMPACT (Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets), and Oncomine cancer gene panel, is 0.959, and 0.989, respectively. The receiver operating characteristic (ROC) curve analysis confirmed that the neural net model has a better prediction performance (Area under the ROC curve (AUC) = 0.992). The use of text-mined co-occurrence features can contextualize each gene. We believe the approach is to evaluate several existing gene panels, and show that we can use part of the gene panel set to predict the remaining genes for cancer discovery.

Deciphering genes associated with diffuse large B-cell lymphoma with lymphomatous effusions: A mutational accumulation scoring approach.

INTRODUCTION: Earlier studies have shown that lymphomatous effusions in patients with diffuse large B-cell lymphoma (DLBCL) are associated with a very poor prognosis, even worse than for non-effusion-associated patients with stage IV disease. We hypothesized that certain genetic abnormalities were associated with lymphomatous effusions, which would help to identify related pathways, oncogenic mechanisms, and therapeutic targets. METHODS: We compared whole-exome sequencing on DLBCL samples involving solid organs (n = 22) and involving effusions (n = 9). We designed a mutational accumulation-based approach to score each gene and used mutation interpreters to identify candidate pathogenic genes associated with lymphomatous effusions. Moreover, we performed gene-set enrichment analysis from a microarray comparison of effusion-associated versus non-effusion-associated DLBCL cases to extract the related pathways. RESULTS: We found that genes involved in identified pathways or with high accumulation scores in the effusion-based DLBCL cases were associated with migration/invasion. We validated expression of 8 selected genes in DLBCL cell lines and clinical samples: MUC4, SLC35G6, TP53BP2, ARAP3, IL13RA1, PDIA4, HDAC1 and MDM2, and validated expression of 3 proteins (MUC4, HDAC1 and MDM2) in an independent cohort of DLBCL cases with (n = 31) and without (n = 20) lymphomatous effusions. We found that overexpression of HDAC1 and MDM2 correlated with the presence of lymphomatous effusions, and HDAC1 overexpression was associated with the poorest prognosis.  CONCLUSION: Our findings suggest that DLBCL associated with lymphomatous effusions may be associated mechanistically with TP53-MDM2 pathway and HDAC-related chromatin remodeling mechanisms.

Automated Radiology Alert System for Pneumothorax Detection on Chest Radiographs Improves Efficiency and Diagnostic Performance.

We aimed to set up an Automated Radiology Alert System (ARAS) for the detection of pneumothorax in chest radiographs by a deep learning model, and to compare its efficiency and diagnostic performance with the existing Manual Radiology Alert System (MRAS) at the tertiary medical center. This study retrospectively collected 1235 chest radiographs with pneumothorax labeling from 2013 to 2019, and 337 chest radiographs with negative findings in 2019 were separated into training and validation datasets for the deep learning model of ARAS. The efficiency before and after using the model was compared in terms of alert time and report time. During parallel running of the two systems from September to October 2020, chest radiographs prospectively acquired in the emergency department with age more than 6 years served as the testing dataset for comparison of diagnostic performance. The efficiency was improved after using the model, with mean alert time improving from 8.45 min to 0.69 min and the mean report time from 2.81 days to 1.59 days. The comparison of the diagnostic performance of both systems using 3739 chest radiographs acquired during parallel running showed that the ARAS was better than the MRAS as assessed in terms of sensitivity (recall), area under receiver operating characteristic curve, and F1 score (0.837 vs. 0.256, 0.914 vs. 0.628, and 0.754 vs. 0.407, respectively), but worse in terms of positive predictive value (PPV) (precision) (0.686 vs. 1.000). This study had successfully designed a deep learning model for pneumothorax detection on chest radiographs and set up an ARAS with improved efficiency and overall diagnostic performance.

WinBinVec: Cancer-Associated Protein-Protein Interaction Extraction and Identification of 20 Various Cancer Types and Metastasis Using Different Deep Learning Models.

Biophysical protein-protein interactions perform dominant roles in the initiation and progression of many cancer-related pathways. A protein-protein interaction might play different roles in diverse cancer types. Hence, prioritizing the PPIs in each cancer type would help detect cancer-associated pathways, find a better understanding of cancer biology, and facilitate drug discovery. Several studies to date have proposed computational methods for extracting the PPI essentiality of different cancer types based on the PPI network. The main drawback of these studies is not using a rich source such as genomics variant data. An amino acid sequence encodes useful information about protein structure and behavior. We represent each amino acid sequence based on its variants/mutations in seven different ways: binary vectors, pathogenicity scores, binding affinity changes upon mutations, gene expression-based network of the interactions, biophysicochemical properties, g-gap dipeptide, and one-hot vectors. Based on these representations, we design and consider seven different deep learning models. Then, we compare the accuracy of these models in predicting 20 different cancer types from the TCGA cohort. WinBinVec is a window-based model that outperforms the other models. Moreover, WinBinVec contains a PPI essentiality module that helps extract the essentiality probability of each PPI for every cancer type. Source code and Data: https://github.com/sabdollahi/WinBinVec.

Comprehensive assessments of germline deletion structural variants reveal the association between prognostic MUC4 and CEP72 deletions and immune response gene expression in colorectal cancer patients.

BACKGROUND: Functional disruptions by large germline genomic structural variants in susceptible genes are known risks for cancer. We used deletion structural variants (DSVs) generated from germline whole-genome sequencing (WGS) and DSV immune-related association tumor microenvironment (TME) to predict cancer risk and prognosis. METHODS: We investigated the contribution of germline DSVs to cancer susceptibility and prognosis by silicon and causal inference models. DSVs in germline WGS data were generated from the blood samples of 192 cancer and 499 non-cancer subjects. Clinical information, including family cancer history (FCH), was obtained from the National Cheng Kung University Hospital and Taiwan Biobank. Ninety-nine colorectal cancer (CRC) patients had immune response gene expression data. We used joint calling tools and an attention-weighted model to build the cancer risk predictive model and identify DSVs in familial cancer. The survival support vector machine (survival-SVM) was used to select prognostic DSVs. RESULTS: We identified 671 DSVs that could predict cancer risk. The area under the curve (AUC) of the receiver operating characteristic curve (ROC) of the attention-weighted model was 0.71. The 3 most frequent DSV genes observed in cancer patients were identified as ADCY9, AURKAPS1, and RAB3GAP2 (p < 0.05). The DSVs in SGSM2 and LHFPL3 were relevant to colorectal cancer. We found a higher incidence of FCH in cancer patients than in non-cancer subjects (p < 0.05). SMYD3 and NKD2DSV genes were associated with cancer patients with FCH (p < 0.05). We identified 65 immune-associated DSV markers for assessing cancer prognosis (p < 0.05). The functional protein of MUC4 DSV gene interacted with MAGE1 expression, according to the STRING database. The causal inference model showed that deleting the CEP72 DSV gene affect the recurrence-free survival (RFS) of IFIT1 expression. CONCLUSIONS: We established an explainable attention-weighted model for cancer risk prediction and used the survival-SVM for prognostic stratification by using germline DSVs and immune gene expression datasets. Comprehensive assessments of germline DSVs can predict the cancer risk and clinical outcome of colon cancer patients.

Precise uncertain significance prediction using latent space matrix factorization models: genomics variant and heterogeneous clinical data-driven approaches.

Several studies to date have proposed different types of interpreters for measuring the degree of pathogenicity of variants. However, in predicting the disease type and disease-gene associations, scholars face two essential challenges, namely the vast number of existing variants and the existence of variants which are recognized as variant of uncertain significance (VUS). To tackle these challenges, we propose algorithms to assign a significance to each gene rather than each variant, describing its degree of pathogenicity. Since the interpreters identified most of the variants as VUS, most of the gene scores were identified as uncertain significance. To predict the uncertain significance scores, we design two matrix factorization-based models: the common latent space model uses genomics variant data as well as heterogeneous clinical data, while the single-matrix factorization model can be used when heterogeneous clinical data are unavailable. We have managed to show that the models successfully predict the uncertain significance scores with low error and high accuracy. Moreover, to evaluate the effectiveness of our novel input features, we train five different multi-label classifiers including a feedforward neural network with the same feature set and show they all achieve high accuracy as the main impact of our approach comes from the features. Availability: The source code is freely available at https://github.com/sabdollahi/CoLaSpSMFM.

Application of an Artificial Intelligence Trilogy to Accelerate Processing of Suspected Patients With SARS-CoV-2 at a Smart Quarantine Station: Observational Study.

BACKGROUND: As the COVID-19 epidemic increases in severity, the burden of quarantine stations outside emergency departments (EDs) at hospitals is increasing daily. To address the high screening workload at quarantine stations, all staff members with medical licenses are required to work shifts in these stations. Therefore, it is necessary to simplify the workflow and decision-making process for physicians and surgeons from all subspecialties. OBJECTIVE: The aim of this paper is to demonstrate how the National Cheng Kung University Hospital artificial intelligence (AI) trilogy of diversion to a smart quarantine station, AI-assisted image interpretation, and a built-in clinical decision-making algorithm improves medical care and reduces quarantine processing times. METHODS: This observational study on the emerging COVID-19 pandemic included 643 patients. An "AI trilogy" of diversion to a smart quarantine station, AI-assisted image interpretation, and a built-in clinical decision-making algorithm on a tablet computer was applied to shorten the quarantine survey process and reduce processing time during the COVID-19 pandemic. RESULTS: The use of the AI trilogy facilitated the processing of suspected cases of COVID-19 with or without symptoms; also, travel, occupation, contact, and clustering histories were obtained with the tablet computer device. A separate AI-mode function that could quickly recognize pulmonary infiltrates on chest x-rays was merged into the smart clinical assisting system (SCAS), and this model was subsequently trained with COVID-19 pneumonia cases from the GitHub open source data set. The detection rates for posteroanterior and anteroposterior chest x-rays were 55/59 (93%) and 5/11 (45%), respectively. The SCAS algorithm was continuously adjusted based on updates to the Taiwan Centers for Disease Control public safety guidelines for faster clinical decision making. Our ex vivo study demonstrated the efficiency of disinfecting the tablet computer surface by wiping it twice with 75% alcohol sanitizer. To further analyze the impact of the AI application in the quarantine station, we subdivided the station group into groups with or without AI. Compared with the conventional ED (n=281), the survey time at the quarantine station (n=1520) was significantly shortened; the median survey time at the ED was 153 minutes (95% CI 108.5-205.0), vs 35 minutes at the quarantine station (95% CI 24-56; P<.001). Furthermore, the use of the AI application in the quarantine station reduced the survey time in the quarantine station; the median survey time without AI was 101 minutes (95% CI 40-153), vs 34 minutes (95% CI 24-53) with AI in the quarantine station (P<.001). CONCLUSIONS: The AI trilogy improved our medical care workflow by shortening the quarantine survey process and reducing the processing time, which is especially important during an emerging infectious disease epidemic.

Automated Recognition of Regional Wall Motion Abnormalities Through Deep Neural Network Interpretation of Transthoracic Echocardiography.

BACKGROUND: Automated interpretation of echocardiography by deep neural networks could support clinical reporting and improve efficiency. Whereas previous studies have evaluated spatial relationships using still frame images, we aimed to train and test a deep neural network for video analysis by combining spatial and temporal information, to automate the recognition of left ventricular regional wall motion abnormalities. METHODS: We collected a series of transthoracic echocardiography examinations performed between July 2017 and April 2018 in 2 tertiary care hospitals. Regional wall abnormalities were defined by experienced physiologists and confirmed by trained cardiologists. First, we developed a 3-dimensional convolutional neural network model for view selection ensuring stringent image quality control. Second, a U-net model segmented images to annotate the location of each left ventricular wall. Third, a final 3-dimensional convolutional neural network model evaluated echocardiographic videos from 4 standard views, before and after segmentation, and calculated a wall motion abnormality confidence level (0-1) for each segment. To evaluate model stability, we performed 5-fold cross-validation and external validation. RESULTS: In a series of 10 638 echocardiograms, our view selection model identified 6454 (61%) examinations with sufficient image quality in all standard views. In this training set, 2740 frames were annotated to develop the segmentation model, which achieved a Dice similarity coefficient of 0.756. External validation was performed in 1756 examinations from an independent hospital. A regional wall motion abnormality was observed in 8.9% and 4.9% in the training and external validation datasets, respectively. The final model recognized regional wall motion abnormalities in the cross-validation and external validation datasets with an area under the receiver operating characteristic curve of 0.912 (95% CI, 0.896-0.928) and 0.891 (95% CI, 0.834-0.948), respectively. In the external validation dataset, the sensitivity was 81.8% (95% CI, 73.8%-88.2%), and specificity was 81.6% (95% CI, 80.4%-82.8%). CONCLUSIONS: In echocardiographic examinations of sufficient image quality, it is feasible for deep neural networks to automate the recognition of regional wall motion abnormalities using temporal and spatial information from moving images. Further investigation is required to optimize model performance and evaluate clinical applications.

Reproducible Machine Learning Methods for Lung Cancer Detection Using Computed Tomography Images: Algorithm Development and Validation.

BACKGROUND: Chest computed tomography (CT) is crucial for the detection of lung cancer, and many automated CT evaluation methods have been proposed. Due to the divergent software dependencies of the reported approaches, the developed methods are rarely compared or reproduced. OBJECTIVE: The goal of the research was to generate reproducible machine learning modules for lung cancer detection and compare the approaches and performances of the award-winning algorithms developed in the Kaggle Data Science Bowl. METHODS: We obtained the source codes of all award-winning solutions of the Kaggle Data Science Bowl Challenge, where participants developed automated CT evaluation methods to detect lung cancer (training set n=1397, public test set n=198, final test set n=506). The performance of the algorithms was evaluated by the log-loss function, and the Spearman correlation coefficient of the performance in the public and final test sets was computed. RESULTS: Most solutions implemented distinct image preprocessing, segmentation, and classification modules. Variants of U-Net, VGGNet, and residual net were commonly used in nodule segmentation, and transfer learning was used in most of the classification algorithms. Substantial performance variations in the public and final test sets were observed (Spearman correlation coefficient = .39 among the top 10 teams). To ensure the reproducibility of results, we generated a Docker container for each of the top solutions. CONCLUSIONS: We compared the award-winning algorithms for lung cancer detection and generated reproducible Docker images for the top solutions. Although convolutional neural networks achieved decent accuracy, there is plenty of room for improvement regarding model generalizability.

Detecting Potential Adverse Drug Reactions Using a Deep Neural Network Model.

BACKGROUND: Adverse drug reactions (ADRs) are common and are the underlying cause of over a million serious injuries and deaths each year. The most familiar method to detect ADRs is relying on spontaneous reports. Unfortunately, the low reporting rate of spontaneous reports is a serious limitation of pharmacovigilance. OBJECTIVE: The objective of this study was to identify a method to detect potential ADRs of drugs automatically using a deep neural network (DNN). METHODS: We designed a DNN model that utilizes the chemical, biological, and biomedical information of drugs to detect ADRs. This model aimed to fulfill two main purposes: identifying the potential ADRs of drugs and predicting the possible ADRs of a new drug. For improving the detection performance, we distributed representations of the target drugs in a vector space to capture the drug relationships using the word-embedding approach to process substantial biomedical literature. Moreover, we built a mapping function to address new drugs that do not appear in the dataset. RESULTS: Using the drug information and the ADRs reported up to 2009, we predicted the ADRs of drugs recorded up to 2012. There were 746 drugs and 232 new drugs, which were only recorded in 2012 with 1325 ADRs. The experimental results showed that the overall performance of our model with mean average precision at top-10 achieved is 0.523 and the rea under the receiver operating characteristic curve (AUC) score achieved is 0.844 for ADR prediction on the dataset. CONCLUSIONS: Our model is effective in identifying the potential ADRs of a drug and the possible ADRs of a new drug. Most importantly, it can detect potential ADRs irrespective of whether they have been reported in the past.

Gene ontology concept recognition using named concept: understanding the various presentations of the gene functions in biomedical literature.

OBJECTIVE: A major challenge in precision medicine is the development of patient-specific genetic biomarkers or drug targets. The firsthand information of the genes associated with the pathologic pathways of interest is buried in the ocean of biomedical literature. Gene ontology concept recognition (GOCR) is a biomedical natural language processing task used to extract and normalize the mentions of gene ontology (GO), the controlled vocabulary for gene functions across many species, from biomedical text. The previous GOCR systems, using either rule-based or machine-learning methods, treated GO concepts as separate terms and did not have an efficient way of sharing the common synonyms among the concepts. MATERIALS AND METHODS: We used the CRAFT corpus in this study. Targeting the compositional structure of the GO, we introduced named concept, the basic conceptual unit which has a conserved name and is used in other complex concepts. Using the named concepts, we separated the GOCR task into dictionary-matching and machine-learning steps. By harvesting the surface names used in the training data, we wildly boosted the synonyms of GO concepts via the connection of the named concepts and then enhanced the capability to recognize more GO concepts in the text. The source code is available athttps://github.com/jeroyang/ncgocr. RESULTS: Named concept gene ontology concept recognizer (NCGOCR) achieved 0.804 precision and 0.715 recall by correct recognition of the non-standard mentions of the GO concepts. DISCUSSION: The lack of consensus on GO naming causes diversity in the GO mentions in biomedical manuscripts. The high performance is owed to the stability of the composing GO concepts and the lack of variance in the spelling of named concepts. CONCLUSION: NCGOCR reduced the arduous work of GO annotation and amended the process of searching for the biomarkers or drug targets, leading to improved biomarker development and greater success in precision medicine.