[Discovering hidden cluster structures in patients with cirrhosis based on laboratory data]

ABSTRACT

Liver cirrhosis was one of the most important liver diseases. Other chronic liver diseases could be lead to liver cirrhosis. Liver cirrhosis could be lead one kind of liver cancers named hepatocellular carcinoma. Cirrhosis in the early stages just by laboratory and imaging testes could be diagnosed. In this study cirrhotic patients were classified based on laboratory symptoms. For this purpose data mining approach has been used in this research. Data mining was an interdisciplinary science that discovers the hidden knowledge in the data. We used K-Means algorithm to categorize the statues of cirrhotic patients. In order to determine the quality of clustering results and to find the best number of clusters, we have used silhouette indices. Our data consists of 410 records which have been collected from Dr. Shariati hospital. The number of features in this study are 0.1 items and sampling were divided into two main groups. At first, we have done clustering based on 21 attributes and the average silhouette was 41 percent. We improved the model, in order to reach a reasonable structure. Finally, based on 7 attributes, a reasonable clustering model was derived. The new model provides 64 percent average silhouette, and based on patients' status, patients are divided into 2 main categories. The risk of HCC in the first cluster is 23 percent and in the second cluster is 14 percent