Hongmei Yan

A multilayer perceptron-based medical decision support system for heart disease diagnosis

The medical diagnosis by nature is a complex and fuzzy cognitive process, and soft computing methods, such as neural networks, have shown great potential to be applied in the development of medical decision support systems (MDSS). In this paper, a multiplayer perceptron-based decision support system is developed to support the diagnosis of heart diseases. The input layer of the system includes 40 input variables, categorized into four groups and then encoded using the proposed coding schemes. The number of nodes in the hidden layer is determined through a cascade learning process. Each of the 5 nodes in the output layer corresponds to one heart disease of interest. In the system, the missing data of a patient are handled using the substituting mean method. Furthermore, an improved back propagation algorithm is used to train the system. A total of 352 medical records collected from the patients suffering from five heart diseases have been used to train and test the system. In particular, three assessment methods, cross validation, holdout and bootstrapping, are applied to assess the generalization of the system. The results show that the proposed MLP-based decision support system can achieve very high diagnosis accuracy (>90%) and comparably small intervals (<5%), proving its usefulness in support of clinic decision process of heart diseases.

Selecting critical clinical features for heart diseases diagnosis with a real-coded genetic algorithm

In clinic, normally a lot of diagnostic features are recorded from a patient for a certain disease. It will be beneficial for the prompt and correct diagnosis of the disease by selecting the important and relevant features and discarding those irrelevant and redundant ones. In this paper, a real-coded genetic algorithm (GA)-based system is proposed to select the critical clinical features essential to the heart diseases diagnosis. The heart disease database used in this study includes 352 cases, and 40 diagnostic features were recorded for each case. Using the proposed genetic algorithm, 24 critical features have been identified, and their corresponding diagnosis weights for each heart disease of interest have been determined. The critical diagnostic features and their clinic meanings are in sound agreement with those used by the physicians in making their clinic decisions.

Development of a decision support system for heart disease diagnosis using multilayer perceptron

In this paper, a computational model based on a multilayer perceptron (MLP) neural network with three layers is employed to develop a decision support system for the diagnosis of five major heart diseases. The input layer of the system includes 38 input variables, extracted from a large number of patient cases. The number of nodes in the hidden layer is determined through a cascade learning process. Each of the 5 nodes in the output layer corresponds to one heart disease of interest. The proposed decision support system is trained using a back propagation algorithm augmented with the momentum term, the adaptive learning rate and the forgetting mechanics. In addition, the missing data are handled using the substituting mean method. The experimental results have shown that the adopted MLP-based decision model can achieve high accuracy level (63.6-82.9%) on the classification of heart diseases, qualifying it as a good decision support system deployable in clinics.

Effect of heartbeat perception on heartbeat evoked potential waves

Early researches found that different heartbeat perceivers have different heartbeat evoked potential (HEP) waves. Two tasks were considered in our experiments to get more details about the differences between good and poor heartbeat perceivers at attention and resting state. Thirty channels of electroencephalogram (EEG) were recorded in 22 subjects, who had been subdivided into good and poor heartbeat perceivers by mental tracking task. Principal component analysis (PCA) was applied to remove cardiac field artifact (CFA) from the HEP. (1) The good heart-beat perceivers showed difference between attention and resting state in the windows from 250 ms to 450 ms after R wave at C3 location and from 100 ms to 300 ms after R wave at C4 location; (2) The difference waveforms between good and poor heartbeat perceivers was a positive waveform at FZ from 220 ms to 340 ms after R wave, which was more significant in attention state. Attention state had more effect on the HEPs of good heartbeat perceivers than that of poor heartbeat perceivers; and perception ability influenced HEPs more strongly in the attention state than in the resting state. 早期研究发现心跳感知能力不同的人其心跳诱发脑电位 (heartbeat evoked potential, HEP) 存在差异。本文设计两个实验以对比研究在关注状态和静息状态下, 心跳感知能力强和心跳感知能力弱的人 HEP 的具体差异。 采集 22 名被试者的30 通道的脑电波, 通过精神跟踪方法将被试者分为心跳感知能力强和感知能力弱两组。 使用主成分分析方法去除掺杂在 HEP 中的心电场伪迹。 (1) 对于心跳感知能力强的被试者, 关注状态和静息状态的 HEP 存在着差异。差异主要体现在C3处心电R波后 250–450 ms 和C4处心电R波后 100–300 ms。 (2) 感知能力强和感知能力弱的被试者的 HEP 差异波主要体现为FZ处心电R波后220–340 ms的一个正波, 在关注状态下: 这种差异更明显。 关注因素可能更容易影响心跳感知能力强的被试者?在关注状态: 感知能力对 HEP 的影响更为明显。

The internet-based knowledge acquisition and management method to construct large-scale distributed medical expert systems

The Internet offers an unprecedented opportunity to construct powerful large-scale medical expert systems (MES). In these systems, a cost-effective medical knowledge acquisition (KA) and management scheme is highly desirable to handle the large quantities of, often conflicting, medical information collected from medical experts in different medical fields and from different geographical regions. In this paper, we demonstrate that a medical KA/management system can be built upon a three-tier distributed client/server architecture. The knowledge in the system is stored/managed in three knowledge bases. The maturity of the medical know-how controls the knowledge flow through these knowledge bases. In addition, to facilitate the knowledge representation and application in these knowledge bases as well as information retrieval across the Internet, an 8-digit numeric coding scheme with a weight value system is proposed. At present, a medical KA and management system based on the proposed method is being tested in clinics. Current results have showed that the method is a viable solution to construct, modify, and expand a distributed MES through the Internet.

SVM-based decision support system for clinic aided tracheal intubation predication with multiple features

During routine anaesthesia, an airway physical examination should be conducted in all patients to estimate whether tracheal intubation is easy or difficult. In clinic, some anaesthetists usually do this by examining single item although most of the specialists agree that full consideration of multiple features of airway physical examination rather than single one would enable anaesthetists to improve the prediction accuracy when encountering a difficult airway. The application of machine learning tools has shown its advantage in medical aided decision. The purpose of this study is to construct a medical decision support system based on support vector machines with 13 physical features for tracheal intubation predication ahead of anaesthesia. A total of 264 medical records collected from patients suffering from a variety of diseases ensure the generalization performance of the decision system. Moreover, the robustness of the proposed system is examined using 4-fold cross-validation method and results show the SVM-based decision support system can achieve average classification accuracy at 90.53%, manifesting its great application prospect of supporting clinic aided diagnosis with full consideration of multiple features of airway physical examination.

Modulation of microsaccade rate by task difficulty revealed through between- and within-trial comparisons.

Microsaccades (MSs) are small eye movements that occur during attempted visual fixation. While most studies concerning MSs focus on their roles in visual processing, some also suggest that the MS rate can be modulated by the amount of mental exertion involved in nonvisual processing. The current study focused on the effects of task difficulty on MS rate in a nonvisual mental arithmetic task. Experiment 1 revealed a general inverse relationship between MS rate and subjective task difficulty. During Experiment 2, three task phases with different requirements were identified: during calculation (between stimulus presentation and response), postcalculation (after reporting an answer), and a control condition (undergoing a matching sequence of events without the need to make a calculation). MS rate was observed to approximately double from the during-calculation phase to the postcalculation phase, and was significantly higher in the control condition compared to postcalculation. Only during calculation was the MS rate generally decreased with greater task difficulty. Our results suggest that the nonvisual cognitive processing can suppress MS rate, and that the extent of such suppression is related to the task difficulty.

Field of Attention for Instantaneous Object Recognition

Background Instantaneous object discrimination and categorization are fundamental cognitive capacities performed with the guidance of visual attention. Visual attention enables selection of a salient object within a limited area of the visual field; we referred to as “field of attention” (FA). Though there is some evidence concerning the spatial extent of object recognition, the following questions still remain unknown: (a) how large is the FA for rapid object categorization, (b) how accuracy of attention is distributed over the FA, and (c) how fast complex objects can be categorized when presented against backgrounds formed by natural scenes. Methodology/Principal Findings To answer these questions, we used a visual perceptual task in which subjects were asked to focus their attention on a point while being required to categorize briefly flashed (20 ms) photographs of natural scenes by indicating whether or not these contained an animal. By measuring the accuracy of categorization at different eccentricities from the fixation point, we were able to determine the spatial extent and the distribution of accuracy over the FA, as well as the speed of categorizing objects using stimulus onset asynchrony (SOA). Our results revealed that subjects are able to rapidly categorize complex natural images within about 0.1 s without eye movement, and showed that the FA for instantaneous image categorization covers a visual field extending 20°×24°, and accuracy was highest (>90%) at the center of FA and declined with increasing eccentricity. Conclusions/Significance In conclusion, human beings are able to categorize complex natural images at a glance over a large extent of the visual field without eye movement.

Machine-learning-based classification of real-time tissue elastography for hepatic fibrosis in patients with chronic hepatitis B.

Hepatic fibrosis is a common middle stage of the pathological processes of chronic liver diseases. Clinical intervention during the early stages of hepatic fibrosis can slow the development of liver cirrhosis and reduce the risk of developing liver cancer. Performing a liver biopsy, the gold standard for viral liver disease management, has drawbacks such as invasiveness and a relatively high sampling error rate. Real-time tissue elastography (RTE), one of the most recently developed technologies, might be promising imaging technology because it is both noninvasive and provides accurate assessments of hepatic fibrosis. However, determining the stage of liver fibrosis from RTE images in a clinic is a challenging task. In this study, in contrast to the previous liver fibrosis index (LFI) method, which predicts the stage of diagnosis using RTE images and multiple regression analysis, we employed four classical classifiers (i.e., Support Vector Machine, Naïve Bayes, Random Forest and K-Nearest Neighbor) to build a decision-support system to improve the hepatitis B stage diagnosis performance. Eleven RTE image features were obtained from 513 subjects who underwent liver biopsies in this multicenter collaborative research. The experimental results showed that the adopted classifiers significantly outperformed the LFI method and that the Random Forest(RF) classifier provided the highest average accuracy among the four machine algorithms. This result suggests that sophisticated machine-learning methods can be powerful tools for evaluating the stage of hepatic fibrosis and show promise for clinical applications.

Where Does the Driver Look? Top-Down-Based Saliency Detection in a Traffic Driving Environment

A traffic driving environment is a complex and dynamically changing scene. When driving, drivers always allocate their attention to the most important and salient areas or targets. Traffic saliency detection, which computes the salient and prior areas or targets in a specific driving environment, is an indispensable part of intelligent transportation systems and could be useful in supporting autonomous driving, traffic sign detection, driving training, car collision warning, and other tasks. Recently, advances in visual attention models have provided substantial progress in describing eye movements over simple stimuli and tasks such as free viewing or visual search. However, to date, there exists no computational framework that can accurately mimic a drivers gaze behavior and saliency detection in a complex traffic driving environment. In this paper, we analyzed the eye-tracking data of 40 subjects consisted of nondrivers and experienced drivers when viewing 100 traffic images. We found that a drivers attention was mostly concentrated on the end of the road in front of the vehicle. We proposed that the vanishing point of the road can be regarded as valuable top-down guidance in a traffic saliency detection model. Subsequently, we build a framework of a classic bottom-up and top-down combined traffic saliency detection model. The results show that our proposed vanishing-point-based top-down model can effectively simulate a drivers attention areas in a driving environment.