医学显微图像分割方法研究进展

医学显微图像分割是医学图像处理中的一个经典难题.针对近年来出现的新方法、新理论,对各种分割方法进行了系统论述,主要包括基于数学形态学方法、神经网络分割、模糊分割、小波分析、遗传算法、统计方法和基于特定模型等方法的图像分割.由于显微图像的复杂性,采用单一方法很难准确分割,故对混合方法也作了一定论述.文中还简要讨论了各种方法的特点和局限性.同时对分割的评价体系也做了简要论述.     

搜索空间逐步缩小的遗传算法用于尿沉渣图像特征选择的研究

基于尿沉渣图像特征选择问题,提出一种新的特征优选方法,首先引入类空间分层分类思想,将多类成分特征集优选问题转化为两类成分的特征集优选问题以减少特征数,从而减少了后续分类器的维数复杂度并提高了优选后的特征集对成分的识别率;针对红细胞和白细胞的特征集优选问题,采用改进的遗传算法进行处理,先根据统计实验结果,锁定待选特征集中形态特征和纹理特征相应的两个可分度最大且相互独立的特征,然后使用基因位逐步锁定技术,结合小生境技术和自适应交叉变异算子,提高了遗传算法的搜索性能;最后,为了提高特征集的优选效果和稳定性,引入"多票投选"机制。就多个尿沉渣成分样本进行验证实验,结果表明,该算法优选的特征集与通过其他方式获得的特征集相比,识别率较高,而且明显减少了后续分类器的维数复杂度。     

一种心电数据压缩算法的研究

利用心电信号的相关性，对ECG进行预处理，再进行差值运算，将得到的差值序列进行DCT变换，从而实现心电数据的压缩。     

Combining clinical judgement and statistical data in expert systems: Over-the-telephone management decisions for critical congenital heart disease in the first month of life

When a new-born baby with congenital heart disease is referred to a regional specialist centre, the transportation management is crucial but must be decided on the basis of clinical information obtained over the telephone. We consider algorithmic and naive statistical approaches to helping in this decision, and on the basis of preliminary results the relative strengths and weaknesses are discussed. A synthesised logical and probabilistic approach appears to have the best potential and could be implemented on hand-held computers.Paper presented at symposium on Computer-assisted Decision Support and Database Management in Anaesthesia, Intensive Care and Cardio-pulmonary Medicine, Rotterdam, 1988.     

投影遗传算法

针对一般遗传算法在求解优化问题的多个优化解时的局限性,结合曲面投影与遗传算法的基本思想,提出了一种基于曲面投影的投影遗传算法用于解决此类问题,数值验证的结果表明了本算法的有效性和实用性。     

基于不完整数据的IHB-LightGBM心脏病预测模型

提出基于不完整数据的IHB-LightGBM(Improved Hyperband-Light Gradient Boosting Machine)心脏病预测模型。首先,在Hyperband算法超参数采样的基础上引入了权重值,并通过蓄水池法按特征权重对其进行排序,从而筛选出最优参数以提高算法的参数寻优能力;其次,针对心脏病数据样本小且属性缺失的问题,使用K近邻算法对不完整数据进行缺失值插补,再将处理得到的完整数据进行归一化,使数据映射至0～1范围内;最后,对LightGBM采用改进后的IHB优化算法进行全局参数寻优,建立IHB-LightGBM心脏病预测模型。使用UCI心脏病数据集进行实验,结果表明IHB算法的参数寻优效果优于贝叶斯、随机搜索等优化算法,IHB-LightGBM模型在各项评价指标也上明显高于随机森林、极端随机树等算法,可以获得更快的预测速度和更高的预测精度。     

Reproducibility of the signal-averaged electrocardiogram using individual lead analysis

The aim of the study was to evaluate the immediate reproducibilityof time domain parameters in the signal averaged electrocardiogramusing a new method for endpoint determination in individualFrank leads. The method is based on a statistical model of theelectrocardiogram (ECG) in which maximum likelihood (ML) estimationis employed. The reproducibility of the ML method was comparedto that of conventional time domain analysis using the vectormagnitude (VM) of Frank leads. Fifty-nine patients were includedin the study and two consecutive ECGs were recorded for signalaveraging. The results showed that the mean of the absolutedifference of the filtered QRS duration (QRSD) between two consecutiverecordings was significantly lower for the ML method than theconventional method when employing 60 H_z highpass filtering(2.1 ± 2.2 ms vs 5.9 ± 10.2 ms, P < 0.05).Moreover, the ML method resulted in a significantly longer QRSDcompared to the VM-based method (P < 0.05). The terminalamplitude of the QRS complex (RMS40) showed a greater variabilitythan the QRSD for both methods, although the ML method was associatedwith a higher reproducibility than the VM method for the 60Hz filter. These findings may contribute to a better identificationof patients at high risk of ventricular arrhythmias. A reductionin the number of measurement errors has important implicationswhen QRS changes are analysed over time.     

Improvement of brain single photon emission tomography (SPET) using transmission data acquisition in a four-head SPET scanner

Attenuation coefficient maps (-maps) are a useful way to compensate for non-uniform attenuation when performing single photon emission tomography (SPET). A new method was developed to record single photon transmission data and a-map for the brain was produced using a four-head SPET scanner. Transmission data were acquired by a gamma camera opposite to a flood radioactive source attached to one of four gamma cameras in the four-head SPET scanner. Attenuation correction was performed using the iterative expectation maximization algorithm and the-map. Phantom studies demonstrated that this method could reconstruct the distribution of radioactivity more accurately than conventional methods, even for a severely non-uniform-map, and could improve the quality of SPET images. Clinical application to technetium-99m hexamethylpropylene amine oxime (HMPAO) brain SPET also demonstrated the usefulness of this method. Thus, this method appears to be promising for improvement in the image quality and quantitative accuracy of brain SPET.This work was presented in part at the World Congress on Medical Physics and Biomedical Engineering, 7–12 July 1991, Kyoto, Japan     

基于变精度粗糙-模糊集模型的诊断知识获取

提出一种基于变精度粗糙-模糊集模型的诊断知识获取算法,利用相似性聚类方法自动获取模糊隶属函数,将连续属性表示成模糊值,通过定义模糊相似关系和模糊相似类给出了变精度粗糙-模糊模型的近似表示,并引入蚁群算法求取模糊相似关系下的属性约简,进行诊断知识的获取。将其应用于精对苯二甲酸生产过程尾氧浓度故障诊断知识获取中,结果表明:该算法可以从故障数据中提取更客观有效的诊断规则,在实际故障诊断中具有很好的应用价值。