CLAHE Performance on Histogram-Based Features for Lymphoma Classification using KNN Algorithm

Immunohistochemistry or IHC is a method for gaining better diagnosis on lymphoma patients. Early diagnosis is needed to enhance their survival rate by giving the right treatment before the cancer spreads to other system organs. However, this method is not affordable in some developing countries, including Indonesia, since it requires advanced tools that are only available at certain high-class hospitals and top public hospitals. Therefore, a machine learning model using H&E (hematoxylin and eosin) stained tissue image was developed to reduce the needs of using IHC. Therefore, only samples classified as lymphoma will undergo the IHC analysis for validation. In this study, a weighted KNN (k-nearest neighbour) model was chosen since it has the performance in classifying datasets to three classes: BL (benign lesion), CA (carcinoma), and LY (lymphoma). Carcinoma was put into consideration because lymphoma can be in the organ as the carcinoma, especially for Non-Hodgkins Lymphoma (NHL). The developed model was analyzing the features on image's histogram in every RGB layer. Image preprocessing using CLAHE (contrast limited adaptive histogram equalization) was studied and succeeded to enhance the model performance with an average accuracy at 85.5% and specificity of LY at 90.3%.