Multi-Level Image Segmentation in Slit-Lamp Images: A Comparison Between two Machine Learning Techniques

Many computer algorithms have been developed, providing an initial aided diagnosis to the medical expertise. Most important previous stage in the automatic classification to grading diseases using images is to obtain a well-segmented region of interest from. Several related research in image classification uses a great number of image processing techniques previous to the classification stage. In this paper, we compare the automatic segmentation based on two leading machine learning techniques: Differential Evolution (DE) and the Self-Organizing Multilayer (SOM) Neural Network (NN) methods. The results are also compared with K-means algorithm for multi-level segmentation from slit-lamp images. Segmented images were obtained relying on a thresholding approach based on fuzzy partitions of the image histogram and a fuzzy entropy measure optimized via a neural process and by the evolutive technique. The resulting approaches were also compared with the classical Shannon entropy.