No abstract

No abstract

No abstract

In Vietnam, nasopharyngeal carcinoma (NPC) is the eighth most common cause of death from cancer. Cell-free Epstein Barr virus DNA (cf-EBV DNA) was reported to be present in almost all NPC patients. However, currently available assays in Vietnam can detect cf-EBV DNA in only 67.6% of NPC patients, thus leaving 32.4% of cancer cases undetected. Therefore, in this study, we aim to develop a highly sensitive quantitative PCR (qPCR) assay that measures the load of cf-EBV DNA for the purpose of early detection of NPC, and then evaluate the sensitivity and the specificity of the developed qPCR assay on the clinical samples. The major methods used in this study include primer/TaqMan probe design, cf-DNA extraction, optimization of qPCR assay and statistical analysis. Using an international standard panel from the Chinese University of HongKong, the linear range of developed qPCR assay is from 50-150,000 copies/ml (R2 = 0.99613) and the detection limit has been shown to be 25 copies/ml. The developed assay could detect cf-EBV DNA with a sensitivity of 96.9% (31/32 NPC patients) and cf-EBV DNA has not been detected in 103 out of 105 healthy controls, which corresponds to a specificity of 98%. Consequently, the performance of the optimal assay has achieved remarkably high sensitivity and specificity. Moreover, the detection limit of our optimal qPCR assay is 25 copies/ml of plasma, which is at least ten times better than other assays tested in recent studies in Vietnam. This developed qPCR assay will also form the basis for further studies in Vietnam and will open many new applications in management of NPC.

Machine learning is expected as a potential aid to support human decision-making in disease prediction. In this study, we determined the utility of various machine learning algorithms in classifying peripheral vestibular (PV) and non-PV diseases through equilibrium function test results. The 1009 patients who had undergone our standardized neuro-otological examinations were recruited. We adopted five supervised machine learning algorithms (Random Forest, AdaBoost, Gradient Boosting, Support Vector Machine, and Logistic Regression). After preprocessing, tuning the best hyperparameters using GridSearchCV, and obtaining the final evaluation using scikit-learn in the testing set, the prediction capability was evaluated by various diagnostic test measures, namely accuracy, F1-score, area under the receiver operating characteristic curve, precision, recall, and Matthews correlation coefficient (MCC). All five algorithms yielded relatively good results with the accuracy of each machine learning algorithm ranging from 76–79%, with the best being the Support Vector Machine classifier. In cases where the predictions of the five models were consistent, the accuracy of the PV diagnosis results was improved with a probability of 83%, whereas it increased to 85% for the non-PV diagnosis results. Increasing the number of patients and optimizing classification methods are warranted to obtain the highest diagnostic accuracy.