Modeling a Data Mining Decision Tree and Propose a New Model for the Diagnosis of Skin Cancer by Immunohistochemical Staining Methods

Abstract

Introduction: New diagnostic methods like immunohistochemistry staining in skin cancer can help the physicians to have more accurate diagnosis. The purpose of this study was to compare a method based on decision tree for differential diagnosis of two kind of skin cancer (Basal cell cancer and Squamous cell cancer) based on the results of staining methods. Method: Sixty skin cancer patients’ data from Malaysia were assessed by two methods of decision tree, CART and CHAID, in data mining and using Clementine 12 and SPSS 19. The results of three staining methods including B-cell lymphoma-2 antibody (BCL2), Galectin-3 (Cytoplasm), and Galectin-3 (Nucleus) were analyzed. The best predictive model for decision tree induction was compared with another researcher-made model based on critical values resulted from Receiver Operating Characteristic (ROC) curve analysis. Results: In final synthetic model, the sensitivity and specificity for Basal Cell Carcinoma (BCC) were 82.1% and 100%, and for Squamous Cell Carcinoma (SCC) were 100% and 82.8%, respectively. The overall accuracy of the model was 90.38% and the positive predictive values (PPV) for SCC and BCC were 82.1% and 100%, and the positive likelihood ratios (PLR) were 5.8 and 5.5 respectively. Conclusion: The decision tree model based on two methods of immunohistochemistry staining in skin cancer, can help in the diagnosis of these malignant disease and provide further studies.

Keywords

• data mining
• decision tree
• immunohistochemistry
• bcl2
• galectin-3
• skin cancer