Detection of medical text semantic similarity based on convolutional neural network.

BACKGROUND: Imaging examinations, such as ultrasonography, magnetic resonance imaging and computed tomography scans, play key roles in healthcare settings. To assess and improve the quality of imaging diagnosis, we need to manually find and compare the pre-existing reports of imaging and pathology examinations which contain overlapping exam body sites from electrical medical records (EMRs). The process of retrieving those reports is time-consuming. In this paper, we propose a convolutional neural network (CNN) based method which can better utilize semantic information contained in report texts to accelerate the retrieving process. METHODS: We included 16,354 imaging and pathology report-pairs from 1926 patients who admitted to Shanghai Tongren Hospital and had ultrasonic examinations between 1st May 2017 and 31st July 2017. We adapted the CNN model to calculate the similarities among the report-pairs to identify target report-pairs with overlapping body sites, and compared the performance with other six conventional models, including keyword mapping, latent semantic analysis (LSA), latent Dirichlet allocation (LDA), Doc2Vec, Siamese long short term memory (LSTM) and a model based on named entity recognition (NER). We also utilized graph embedding method to enhance the word representation by capturing the semantic relations information from medical ontologies. Additionally, we used LIME algorithm to identify which features (or words) are decisive for the prediction results and improved the model interpretability. RESULTS: Experiment results showed that our CNN model gained significant improvement compared to all other conventional models on area under the receiver operating characteristic (AUROC), precision, recall and F1-score in our test dataset. The AUROC of our CNN models gained approximately 3-7% improvement. The AUROC of CNN model with graph-embedding and ontology based medical concept vectors was 0.8% higher than the model with randomly initialized vectors and 1.5% higher than the one with pre-trained word vectors. CONCLUSION: Our study demonstrates that CNN model with pre-trained medical concept vectors could accurately identify target report-pairs with overlapping body sites and potentially accelerate the retrieving process for imaging diagnosis quality measurement.

Integrated transcriptomic response to cardiac chronic hypoxia: translation regulators and response to stress in cell survival.

Complementary deoxyribonucleic acid microarray data from 36 mice subjected for 1, 2, or 4 weeks of their early life to normal atmospheric conditions (normoxia) or chronic intermittent (CIH) or constant (CCH) hypoxia were analyzed to extract organizational principles of the developing heart transcriptome and determine the integrated response to oxygen deprivation. Although both CCH and CIH regulated numerous genes involved in a wide diversity of processes, the changes in maturational profile, expression stability, and coordination were vastly different between the two treatments, indicating the activation of distinct regulatory mechanisms of gene transcription. The analysis focused on the main regulators of translation and response to stress because of their role in the cardiac hypertrophy and cell survival in hypoxia. On average, the expression of each heart gene was tied to the expression of about 20% of other genes in normoxia but to only 8% in CCH and 9% in CIH, indicating a strong decoupling effect of hypoxia. In contrast to the general tendency, the interlinkages among components of the translational machinery and response to stress increased significantly in CIH and much more in CCH, suggesting a coordinated response to the hypoxic stress. Moreover, the transcriptomic networks were profoundly and differently remodeled by CCH and CIH.

Transcriptomic changes in developing kidney exposed to chronic hypoxia.

cDNA arrays compared gene expression in kidneys of neonatal mice subjected to 1, 2, and 4 weeks of chronic constant (CCH) or intermittent (CIH) hypoxia with normoxic littermates. Five to twenty percent of genes were regulated in each condition, with greater changes in CCH. Up-regulation of 42% of the solute carriers after 1 week of CCH suggests a strong activation of pH controlling pathways. Significant reduction in expression change of genes important in growth, development, and aging as a function of time indicates reduced maturation rate in CIH and CCH treatments. Regulated genes showed gender dependence in CCH, being higher in females than males at 1 week and higher in males than females thereafter. Transcriptional control was enhanced in CCH but not in CIH. Thus, CCH and CIH both alter gene expression and retard maturation with the more profound changes occurring in CCH than in CIH.

Gene expression and phenotypic characterization of mouse heart after chronic constant or intermittent hypoxia.

Chronic constant hypoxia (CCH), such as in pulmonary diseases or high altitude, and chronic intermittent hypoxia (CIH), such as in sleep apnea, can lead to major changes in the heart. Molecular mechanisms underlying these cardiac alterations are not well understood. We hypothesized that changes in gene expression could help to delineate such mechanisms. The current study used a neonatal mouse model in CCH or CIH combined with cDNA microarrays to determine changes in gene expression in the CCH or CIH mouse heart. Both CCH and CIH induced substantial alterations in gene expression. In addition, a robust right ventricular hypertrophy and cardiac enlargement was found in CCH- but not in CIH-treated mouse heart. On one hand, upregulation in RNA and protein levels of eukaryotic translation initiation factor-2alpha and -4E (eIF-2alpha and eIF-4E) was found in CCH, whereas eIF-4E was downregulated in 1- and 2-wk CIH, suggesting that eIF-4E is likely to play an important role in the cardiac hypertrophy observed in CCH-treated mice. On the other hand, the specific downregulation of heart development-related genes (e.g., notch gene homolog-1, MAD homolog-4) and the upregulation of proteolysis genes (e.g., calpain-5) in the CIH heart can explain the lack of hypertrophy in CIH. Interestingly, apoptosis was enhanced in CCH but not CIH, and this was correlated with an upregulation of proapoptotic genes and downregulation of anti-apoptotic genes in CCH. In summary, our results indicate that 1) the pattern of gene response to CCH is different from that of CIH in mouse heart, and 2) the identified expression differences in certain gene groups are helpful in dissecting mechanisms responsible for phenotypes observed.

Expression of Drosophila trehalose-phosphate synthase in HEK-293 cells increases hypoxia tolerance.

Increasing hypoxia tolerance in mammalian cells is potentially of major importance, but it has not been feasible thus far. The disaccharide trehalose, which accumulates dramatically during heat shock, enhances thermotolerance and reduces aggregation of denatured proteins. Previous studies from our laboratory showed that over-expression of Drosophila trehalose-phosphate synthase (dtps1) increases the trehalose level and anoxia tolerance in flies. To determine whether trehalose can protect against anoxic injury in mammalian cells, we transfected the dtps1 gene into human HEK-293 cells using the recombinant plasmid pcDNA3.1(-)-dtps1 and obtained more than 20 stable cell strains. Glucose starvation in culture showed that HEK-293 cells transfected with pcDNA3.1(-)-dtps1 (HEK-dtps1) do not metabolize intracellular trehalose, and, interestingly, these cells accumulated intracellular trehalose during hypoxic exposure. In contrast to HEK-293 cells transfected with pcDNA3.1(-) (HEK-v), cells with trehalose were more resistant to low oxygen stress (1% O2). To elucidate how trehalose protects cells from anoxic injury, we assayed protein solubility and the amount of ubiquitinated proteins. There was three times more insoluble protein in HEK-v than in HEK-dtps1 after 3 days of exposure to low O2. The amount of Na+-K+ ATPase present in the insoluble proteins dramatically increased in HEK-v cells after 2 and 3 days of exposure, whereas there was no significant change in HEK-dtps1 cells. Ubiquitinated proteins increased dramatically in HEK-v cells after 2 and 3 days of exposure but not in HEK-dtps1 cells over the same period. Our results indicate that increased trehalose in mammalian cells following transfection by the Drosophila tps1 gene protects cells from hypoxic injury. The mechanism of this protection is likely related to a decrease in protein denaturation, through protein-trehalose interactions, resulting in enhanced cellular recovery from hypoxic stress.