SOM-Like Neural Network and Differential Evolution for Multi-level Image Segmentation and Classification in Slit-Lamp Images

A nuclear cataract is a type of disease of the eye that affects a considerable part of the human population at an advanced age. Due to the high demand for clinical services, computer algorithms based on artificial intelligence have emerged, providing acceptable aided diagnostics to the medical field. However, several challenges are yet to be overcome. For instance, a well-segmented image of the region of interest could prove valuable at a previous stage in the automatic classification of this disease. A great variety of research in image classification uses several image processing techniques before the classification stage. In this paper, we explore the automatic segmentation based on two leading techniques, namely, a Self-Organizing Multilayer (SOM) Neural Network (NN) and Differential Evolution (DE) methods. Specifically, the fuzzy entropy measure used here is optimized via a neural process, and by using the evolutive technique, optimal thresholds of the images are obtained. The experimental part shows significant results in getting a useful automatic segmentation of the medical images. In this extended version, we have implemented the use of a Multilayer Perceptron, a classifier that proves the usefulness of the segmented images.