Zhaohui Liang

Deep learning for healthcare decision making with EMRs

Computer aid technology is widely applied in decision-making and outcome assessment of healthcare delivery, in which modeling knowledge and expert experience is technically important. However, the conventional rule-based models are incapable of capturing the underlying knowledge because they are incapable of simulating the complexity of human brains and highly rely on feature representation of problem domains. Thus we attempt to apply a deep model to overcome this weakness. The deep model can simulate the thinking procedure of human and combine feature representation and learning in a unified model. A modified version of convolutional deep belief networks is used as an effective training method for large-scale data sets. Then it is tested by two instances: a dataset on hypertension retrieved from a HIS system, and a dataset on Chinese medical diagnosis and treatment prescription from a manual converted electronic medical record (EMR) database. The experimental results indicate that the proposed deep model is able to reveal previously unknown concepts and performs much better than the conventional shallow models.

CNN-based image analysis for malaria diagnosis

Malaria is a major global health threat. The standard way of diagnosing malaria is by visually examining blood smears for parasite-infected red blood cells under the microscope by qualified technicians. This method is inefficient and the diagnosis depends on the experience and the knowledge of the person doing the examination. Automatic image recognition technologies based on machine learning have been applied to malaria blood smears for diagnosis before. However, the practical performance has not been sufficient so far. This study proposes a new and robust machine learning model based on a convolutional neural network (CNN) to automatically classify single cells in thin blood smears on standard microscope slides as either infected or uninfected. In a ten-fold cross-validation based on 27,578 single cell images, the average accuracy of our new 16-layer CNN model is 97.37%. A transfer learning model only achieves 91.99% on the same images. The CNN model shows superiority over the transfer learning model in all performance indicators such as sensitivity (96.99% vs 89.00%), specificity (97.75% vs 94.98%), precision (97.73% vs 95.12%), F1 score (97.36% vs 90.24%), and Matthews correlation coefficient (94.75% vs 85.25%).

A kernel-decision tree based algorithm for outcome prediction on acupuncture for neck pain: A new method for interim analysis

Neck pain is a common disorder in modern society as the result of changes in working and life style. Acupuncture is a traditional treatment of Chinese medicine for neck pain, whose therapeutic mechanism follows the classic knowledge and understanding of Chinese medicine. Syndrome-based diagnosis and treatment is a significant feature of Chinese medicine, and guides the practice of acupuncture. In the treatment of neck pain, acupuncture provides a standard prescription whose effect is support by latest multi-center RCTs. However, the potential difference of its effectiveness in different syndrome types is challenged due to small sample size and limits of statistical power. In our study, we apply the machine learning methods to a data set of the outcomes of a multi-center RCT clinical trial, which consists of demographical information and efficacy outcomes. A decision tree with kernel mapping was applied as the main algorithm to discover the underlying relationship and difference between clinical outcomes among different syndrome types, and to predict its tendency in trials with larger sample size. Kernel function is used to map the input data items to a feature space with better representation, which yields a smooth KNN classification boundary. Non-Dominated Sort (NDS) is used to obtain an optimal order of the three efficacy outcomes from a small sample at the beginning. Then the proposed method was tested with the clinical data from a large sample from a multi-center RCT conducted from 2006 to 2010. The result shows the proposed algorithm is capable of discovering the underlying difference among different syndrome types and feasible to predict the effective tendency in clinical trials of large sample. Therefore, it provides a potential solution for interim analysis of clinical trials, which overcomes the limitation of conventional statistical methods.

A clinical outcome evaluation model with local sample selection: A study on efficacy of acupuncture for cervical spondylosis

Local learning is a special learning framework that considers training samples located in a small region concentric of the query sample. In many applications the concept label of query sample can be evaluated effectively only by similar training samples, such as the famous K-nearest neighbors (KNN) classifier. The metric of locality or similarity is essential in local learning, which is often application oriented and implied in local geometry of input space. In this paper, we propose to apply local learning to the task of outcome assessment and evaluation on acupuncture for cervical spondylosis (CS) in a multi-center clinical trial. The analytic data are measures of three questionnaires which are recognized tools for subjective patient-reported outcomes (PROs) evaluation. We propose a similarity evaluation method based on both Euclidean distance and the therapy effect of recent records. A Non-Dominated Sort (NDS) based methods is applied to obtain a ranking of therapy effect. A WEKA implementation decision tree classifier is applied as the main learner in our work, with comparison to two base line methods. The result shows that the proposed local learning method dramatically outperforms the global version in both classification accuracy and computational costs.

Discovery of the relations between genetic polymorphism and adverse drug reactions

The genetic polymorphism of Cytochrome P450 (CYP 450) is considered as one of the main causes for adverse drug reactions (ADRs). In order to explore the latent correlations between ADRs and the genetic polymorphism, a new model is proposed in which both the inputs of the genetic locuses (i.e.CYP2D6*2, CYP2D6*10, CYP2D6*14, CYP1A2*1C and CYP1A2*1F) and occurrence as probabilistic distribution. A generative model is proposed to describe the joint distributions of occurrence of ADRs and the diversity of genetic sub-types of the input variables. The new algorithm is developed based on Generative Stochastic Networks (GSN) model. A Markov chain from a training data set is applied for the learning as a transition operator to simulate a probabilistic distribution. The transition distribution is conditional on the previous step of the chain thus it is able to perform learning at a much lower cost than the conventional maximal likelihood method. The experiment results show that the newly algorithm is more effective than the available conventional methods.

Deep generative learning for automated EHR diagnosis of traditional Chinese medicine

BACKGROUND Computer-aided medical decision-making (CAMDM) is the method to utilize massive EMR data as both empirical and evidence support for the decision procedure of healthcare activities. Well-developed information infrastructure, such as hospital information systems and disease surveillance systems, provides abundant data for CAMDM. However, the complexity of EMR data with abstract medical knowledge makes the conventional model incompetent for the analysis. Thus a deep belief networks (DBN) based model is proposed to simulate the information analysis and decision-making procedure in medical practice. The purpose of this paper is to evaluate a deep learning architecture as an effective solution for CAMDM. METHODS A two-step model is applied in our study. At the first step, an optimized seven-layer deep belief network (DBN) is applied as an unsupervised learning algorithm to perform model training to acquire feature representation. Then a support vector machine model is adopted to DBN at the second step of the supervised learning. There are two data sets used in the experiments. One is a plain text data set indexed by medical experts. The other is a structured dataset on primary hypertension. The data are randomly divided to generate the training set for the unsupervised learning and the testing set for the supervised learning. The model performance is evaluated by the statistics of mean and variance, the average precision and coverage on the data sets. Two conventional shallow models (support vector machine / SVM and decision tree / DT) are applied as the comparisons to show the superiority of our proposed approach. RESULTS The deep learning (DBN + SVM) model outperforms simple SVM and DT on two data sets in terms of all the evaluation measures, which confirms our motivation that the deep model is good at capturing the key features with less dependence when the index is built up by manpower. CONCLUSIONS Our study shows the two-step deep learning model achieves high performance for medical information retrieval over the conventional shallow models. It is able to capture the features of both plain text and the highly-structured database of EMR data. The performance of the deep model is superior to the conventional shallow learning models such as SVM and DT. It is an appropriate knowledge-learning model for information retrieval of EMR system. Therefore, deep learning provides a good solution to improve the performance of CAMDM systems.

Fast Screening Technology for Drug Emergency Management: Predicting Suspicious SNPs for ADR with Information Theory-based Models

OBJECTIVE The genetic polymorphism of Cytochrome P450 (CYP 450) is considered as one of the main causes for adverse drug reactions (ADRs). In order to explore the latent correlations between ADRs and potentially corresponding single-nucleotide polymorphism (SNPs) in CYP450, three algorithms based on information theory are used as the main method to predict the possible relation. METHODS The study uses a retrospective case-control study to explore the potential relation of ADRs to specific genomic locations and single-nucleotide polymorphism (SNP). The genomic data collected from 53 healthy volunteers are applied for the analysis, another group of genomic data collected from 30 healthy volunteers excluded from the study are used as the control group. The SNPs respective on five loci of CYP2D6*2,*10,*14 and CYP1A2*1C, *1F are detected by the Applied Biosystem 3130xl. The raw data is processed by ChromasPro to detect the specific alleles on the above loci from each sample. The secondary data are reorganized and processed by R combined with the reports of ADRs from clinical reports. Three information theory based algorithms are implemented for the screening task: JMI, CMIM, and mRMR. If a SNP is selected by more than two algorithms, we are confident to conclude that it is related to the corresponding ADR. The selection results are compared with the control decision tree + LASSO regression model. RESULTS In the study group where ADRs occur, 10 SNPs are considered relevant to the occurrence of a specific ADR by the combined information theory model. In comparison, only 5 SNPs are considered relevant to a specific ADR by the decision tree + LASSO regression model. In addition, the new method detects more relevant pairs of SNP and ADR which are affected by both SNP and dosage. This implies that the new information theory based model is effective to discover correlations of ADRs and CYP 450 SNPs and is helpful in predicting the potential vulnerable genotype for some ADRs. CONCLUSION The newly proposed information theory based model has superiority performance in detecting the relation between SNP and ADR compared to the decision tree + LASSO regression model. The new model is more sensitive to detect ADRs compared to the old method, while the old method is more reliable. Therefore, the selection criteria for selecting algorithms should depend on the pragmatic needs.

Research on the thought of needle-medicine of mutual reinforcement school

This paper firstly define the concept of Combined Acupuncture with medicine, to explore the theoretical basis of acupuncture and drugs, Analyzed from the following aspects: comparative analysis of the advantages and disadvantages of acupuncture and Chinese Medicine, the way and method of combination, he current research progress, study on the mechanism of acupuncture combined with medicine. The objective is to grasp the characteristics and advantages of treatment of the acupuncture and Chinese herbal medicine, to explore organic combination of the two modes of treatment, to find the optimal treatment plan, summary of the law of combination of acupuncture and medicine clinical application. Give full play to the advantages of acupuncture and medicine complementary.

Acupoint injection combined with acupuncture for insomnia with cardiacsplenic asthenia: A research protocol for clinical trial

Insomnia is common psychophysiogical disorder, characterized by sleeping difficultly or lightly, restless sleep and early awakening, usually with a series of disease, such as neurasthenia, anxiety, depression and so on. Nowadays acupuncture is a complementary therapy for insomnia which is regarded as one of most effective, widely used and well-accepted method. However, acupuncture is usually used independently. In this paper, we present a research protocol designed for a parallel, randomized, controlled trial to evaluate the effect of the acupoint injection combined with acupuncture treatment for insomnia with cardiac-splenic asthenia. In our study, the objective is to evaluate the clinical effect of the acupoint injection combined with acupuncture treatment compared with the acupuncture used independently.

Moxibustion at Da-zhui acupoint for sensorineural tinnitus with the pattern about deficiency of qi-and-blood: A study protocol for clinical trial

Sensorineural tinnitus is the chief pattern of hearing disorders characterized by hearing loss, sleep disturbances, inattention, upset, anxiety and depression, etc. Nowadays traditional Chinese medicine such as acupuncture is considered as a widely-used and complementary therapy for tinnitus. However, most of the classic acupuncture is only taken by acupoints while moxibustion is another effective and well - accepted method. In this paper, we prepare a study protocol for sensorineural tinnitus with the pattern about deficiency of qi-and-blood, and attempt to provide a scientific method for further study combining disease with syndrome on tinnitus disorder. In our study, the objective is to evaluate the clinical effect of the moxibustion at Da-zhui acupoints compared with the classic acupuncture treatment measured mainly by the Tinnitus Handicap Inventory (THI) score and audiometry examination results.