Performance of feature selection methods for machine learning based automatic malarial cell recognition in wholeslide images
Date of Award
Master of Science in Engineering (MSE)
Electrical and Computer Engineering
W. David Pan
B. Earl Wells
Vector analysis., Analytic mechanics., Support vector machines., Bioinformatics., Medical informatics.
This work is on investigation of automated detection of malaria parasites in wholeslide images of peripheral blood smears. We compared and analyzed several feature-selection methods for optimization of machine learning using support vector machine (SVM). To this end, cell samples were extracted from the wholeslide image and feature extraction was performed on these samples. In order to identify the best features, a host of feature selection methods were tested. For each set of selected features, the SVM was trained and tuned for optimal model selection using cross-validation and grid-search. Subsequently, test samples were classified using the model. Each feature selection method was assessed in terms of various performance measures. It was concluded that feature selection based on Kullback-Leibler Distance yielded the best results.
Muralidharan., Vishnu, "Performance of feature selection methods for machine learning based automatic malarial cell recognition in wholeslide images" (2016). Theses. 181.