Narrative reports in medical records contain abundant clinical information that may be converted into structured data for managing patient information and predicting trends in diseases. Though various rule-based and machine-learning methods are available in electronic medical records (EMRs), a few works have explored the hybrid methods in extracting information from the Chinese EMRs. In this paper, we developed a novel hybrid approach which integrates the rules and bidirectional long short-term memory with a conditional random field layer (BiLSTM-CRF) model to extract clinical entities and attributes. A corpus of 1509 electronic notes (discharge summaries and operation notes) was annotated. Annotation from three clinicians was reconciled to form a gold standard dataset. The performance of our method was assessed by calculating the precision, recall, and F-measure for two boundary matching strategies. The experimental results demonstrate the effectiveness of our method in clinical information extraction from the Chinese EMRs.

Purpose: Preoperative evaluation of lymph node metastasis (LNM) is the basis of personalized treatment of locally advanced gastric cancer (LAGC). We aim to develop and evaluate CT-based model using deep learning features to preoperatively predict LNM in LAGC.Methods: A combined size of 523 patients who had pathologically confirmed LAGC were retrospectively collected between August 2012 and July 2019 from our hospital. Five pre-trained convolutional neural networks were exploited to extract deep learning features from pretreatment CT images. And the support vector machine (SVM) was employed as the classifier. We assessed the performance using the area under the receiver operating characteristics curve (AUC) and selected an optimal model, which was compared with a radiomics model developed from the training cohort. A clinical model was built with clinical factors only for baseline comparison.Results: The optimal model with features extracted from ResNet yielded better performance with AUC of 0.796 [95% confidence interval (95% CI), 0.715-0.865] and accuracy of 75.2% (95% CI, 67.2%-81.5%) in the testing cohort, compared with 0.704 (0.625-0.783) and 61.8% (54.5%-69.9%) for the radiomics model. The predictive performance of all the radiological models were significantly better than the clinical model.Conclusion: The novel and noninvasive deep learning approach could provide efficient and accurate prediction of lymph node metastasis in LAGC, and benefit clinical decision making of therapeutic strategy.

Intrahepatic cholangiocarcinoma (iCCA) and other subtypes of primary liver cancer (PLC) have overlapping clinical manifestations and radiological characteristics. The objective of this study was to evaluate the efficacy of deep learning (DL) radiomics analysis, performed using computed tomography (CT) and magnetic resonance imaging (MRI), in diagnosing iCCA within PLC. 178 pathologically confirmed PLC patients (training cohort: test cohort = 124: 54) who underwent both CT and MRI examinations was enrolled. Univariate and multivariate analysis was used to identify the significant factors of radiological findings for diagnosing iCCA. DL radiomics analysis was applied to CT and MRI images, respectively. We constructed and evaluated six distinct models: CT DL radiomics (DLRS^CT), CT radiological (R^CT), CT DL radiomics-radiological (DLRR^CT), MRI DL radiomics (DLRS^MRI), MRI radiological (R^MRI) and MRI DL radiomics-radiological (DLRR^MRI). To further explore the diagnostic and predictive value of a cross-modal approach, we developed a fused model that combined DLRR^CT and DLRR^MRI. Receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA) were employed to compare the performance of different models. MRI-based models demonstrated a superior predictive performance than CT-based models in test cohort (AUCs of MRI vs. CT: DLRR, 0.923 vs. 0.880, P = 0.521; DLRS, 0.875 vs. 0.867, P = 0.922; R, 0.859 vs. 0.840, P = 0.808). The CT-MRI cross-modal model yielded the highest AUC of 0.994 and 0.937 in training and test cohorts, respectively. CT- and MRI-based DL radiomics analyses exhibited good performance in diagnosing iCCA, and the CT-MRI cross-modal model may have significant clinical implications on detection of liver malignancies.

Blood cultures are often performed to detect patients who has a serious illness without infections and patients with bloodstream infections. Early positive blood culture prediction is important, as bloodstream infections may cause inflammation of the body, even organ failure or death. However, existing work mainly adopts statistical models with laboratory indicators, and fails to make full use of textual description information from EHRs. We study the problem of positive blood culture prediction by using neural network model. Specifically, we first construct dataset from raw EHRs. Then we propose a hybrid neural network which incorporates attention based Bi-directional Long Short-Term Memory and Autoencoder networks to fully capture the information in EHRs. In order to evaluate the proposed method, we constructe a dataset which consists of totally 5963 patients who had one or more blood cultures tests during hospitalization. Experimental results show that the proposed neural model gets 91.23% F-measure for this task. The comparison results of different models demonstrated the effectiveness of our model. The proposed model outperformed traditional statistical models.

This study aimed to investigate the usefulness of CT-based deep learning radiomics analysis (DLRA) for preoperatively differentiating Lauren classification in gastric cancer (GC) patients and explore the tumor microenvironment. 578 patients were recruited from January 2015 to June 2024, and divided into the training cohort (n = 311), the internal validation cohort (n = 132), and the external validation cohort (n = 135). Clinical characteristics were collected. Radiomics features were extracted from CT images at arterial phase (AP) and venous phase (VP). A radiomics nomogram incorporating radiomics signature and clinical information was built for distinguishing Lauren classification, and its discrimination, calibration, and clinical usefulness were evaluated. RNA sequencing data from The Cancer Imaging Archive database were used to perform transcriptomics analysis. The nomogram incorporating the two radiomics signatures and clinical characteristics exhibited good discrimination of Lauren classification on all cohorts [overall C-indexes 0.815 (95 % CI: 0.739–0.869) in the training cohort, 0.785 (95 % CI: 0.702–0.834) in the internal validation cohort, 0.756 (95 % CI: 0.685–0.816) in the external validation cohort]. It outperformed the clinical model in predictive ability. The calibration and decision curve substantiated the model's excellent fitness and clinical applicability. Further, transcriptomics analysis showed that the differentially expressed genes of different Lauren types were mainly enriched in pathways related to cell contraction and migration, and the infiltration degree of various immune cells was also significantly different. DLRA effectively differentiated Lauren classification in GC, and our analysis of transcriptomic data across different Lauren subtypes revealed the heterogeneity within the GC microenvironment.

Background: Named entity recognition (NER) on Chinese electronic medical/healthcare records has attracted significantly attentions as it can be applied to building applications to understand these records. Most previous methods have been purely data-driven, requiring high-quality and large-scale labeled medical data. However, labeled data is expensive to obtain, and these data-driven methods are difficult to handle rare and unseen entities. Methods: To tackle these problems, this study presents a novel multi-task deep neural network model for Chinese NER in the medical domain. We incorporate dictionary features into neural networks, and a general secondary named entity segmentation is used as auxiliary task to improve the performance of the primary task of named entity recognition. Results: In order to evaluate the proposed method, we compare it with other currently popular methods, on three benchmark datasets. Two of the datasets are publicly available, and the other one is constructed by us. Experimental results show that the proposed model achieves 91.07% average f-measure on the two public datasets and 87.05% f-measure on private dataset. Conclusions: The comparison results of different models demonstrated the effectiveness of our model. The proposed model outperformed traditional statistical models.