Cui Jie, Loewy John, Kendall Edward J
Institute of Biomaterials and Biomedical Engineering, University of Toronto, Toronto, ON M5S 3G9, Canada.
IEEE Trans Biomed Eng. 2006 May;53(5):800-9. doi: 10.1109/TBME.2006.872813.
Analysis of synovial fluid by infrared (IR) clinical chemistry requires expert interpretation and is susceptible to subjective error. The application of automated pattern recognition (APR) may enhance the utility of IR analysis. Here, we describe an APR method based on the fuzzy C-means cluster adaptive wavelet (FCMC-AW) algorithm, which consists of two parts: one is a FCMC using the features from an M-band feature extractor adopting the adaptive wavelet algorithm and the second is a Bayesian classifier using the membership matrix generated by the FCMC. A FCMC-cross-validated quadratic probability measure (FCMC-CVQPM) criterion is used under the assumption that the class probability density is equal to the value of the membership matrix. Therefore, both values of posterior probabilities and selection criterion MFQ can be obtained through the membership matrix. The distinctive advantage of this method is that it provides not only the 'hard' classification of a new pattern, but also the confidence of this classification, which is reflected by the membership matrix.
通过红外（IR）临床化学分析滑液需要专家解读，且易受主观误差影响。自动模式识别（APR）的应用可能会提高红外分析的效用。在此，我们描述一种基于模糊C均值聚类自适应小波（FCMC-AW）算法的APR方法，该方法由两部分组成：一部分是使用采用自适应小波算法的M波段特征提取器的特征的FCMC，另一部分是使用由FCMC生成的隶属度矩阵的贝叶斯分类器。在类概率密度等于隶属度矩阵值的假设下，使用FCMC交叉验证二次概率度量（FCMC-CVQPM）准则。因此，后验概率值和选择标准MFQ都可以通过隶属度矩阵获得。该方法的独特优势在于，它不仅提供新模式的“硬”分类，还提供这种分类的置信度，这由隶属度矩阵体现。
