A precise classifier for the substances in urinary sediment images based on neural networks and fuzzy reasoning
N.F. Zhen, K. Taniguchi, Satoru Watanabe, Yuya Nakano, H. Nakamoto · 2002
A new method is proposed to classify the substances in urinary sediment images based on the combination of neural networks and fuzzy reasoning. First, the features of the normal substance for each one are collected. Second, based on these features, each kind of normal substance can be classified from all substances in the urinary sediment images by using the BP-NN. One structure of the NN is designed for each substance and all of them are trained separately. After that, if abnormal substances cannot be separated properly by NN, they are classified further by using fuzzy reasoning. A database is created for storing the different types of substances which include normal and abnormal substances used as standard patterns. Then the similarity degrees between the standard patterns and tested substances are calculated. Furthermore, other features such as the texture are used for creating an "If-then" knowledge base based on the experiences of expert. Finally, the knowledge base for each substance is applied to evaluate the abnormal substance by using fuzzy reasoning. As a result, the accuracy of the automatic classification is improved.