Application of two step clustering in pathological image analysis

  Objective   To investigate the potential of two step cluster ing analysis in patho logicalimage analysis.

  Methods   Two step clustering was used to cluster normal, low-grade squamous intraepithelial lesion, high-grade squamous intraepit helial lesion cervical cells with 51 char acters.Firstly, the cases were clustered intomany small sub-clusters and then the sub-clusters were clustered into the desired number of clusters.The number of clusters was automatically determined by Bayesian Information Criteria with log-likelihood distance measure.

  Results   Classification accuracy of normal, low-grade squamous intr aepithelial lesion and high-grade squamous intraepithelial lesion cervical cells was 98.0%, 96.1% and 100%, respectively.

  Conclusion   With the high classification accuracy, the two step clustering can measure the variables.significance, which could provide useful information for image analysis.