Extract Knowledge from Site-sampled Data Sets and Fused Hierarchical Neural Networks for Detecting Cardiovascular Diseases

For purpose of detecting cardiovascular diseases (CVDs) hierarchically via hemodynamic parameters (HDPs) derived from sphygmogram, a fused hierarchical neural networks (FHNNs) scheme is proposed, which provides a noninvasive way to detect CVDs. To deduce conclusion via FHNNs, method of variance analysis is used to categorize HDPs. The categorized HDP sets are then inhaled by different sub neural networks (sub-NNs) of FHNNs according to importance and relevance of sets, and outputs of former sub-NNs are inhaled as inputs of latter sub-NNs, so that FHNNs have higher overall accuracy, especially in dealing with mixed CVDs. The sub-NNs are implemented by probabilistic neural networks (PNNs) or learning vector quantization networks (LVQ) due to their diverse advantages in dealing with different HDP categories. The preliminary testing results on site sampled data sets prove that the proposed method is promising for detecting CVDs.