A neuro-fuzzy system for automated detection and classification of human intestinal parasites

Abstract

Abstract Background and objective Human intestinal parasites are a major public health concern in tropical countries. The most reliable diagnosis of these parasites relies on the visual analysis of stool specimens. However, this method is time consuming, tedious, and prone to diagnosis error. Hence, the aim of this work is the automatic analysis of microscopic images for the classification of intestinal parasites. A combination of a fuzzy system and an artificial neural network leads to an artificial intelligence system appropriate for this task. Methods The approach is based on segmentation and training of a classifier. The input to the system is a microscopic image of a given stool sample with parasites. The parasite is firstly localized by the circular Hough transform and secondly, the distance regularized level set evolution is automatically initialized for segmentation. From the extracted parasite, we determined the histogram oriented region with feature vectors being of high interest. The dimension of feature vectors is reduced using linear discriminant analysis and is considered as input to the classifier. Finally, the neuro-fuzzy classifier is trained according to a speeded up scaled conjugate gradient algorithm. Results The proposed scheme has been applied for recognition and classification of twenty human intestinal parasites. The results demonstrate satisfactory classification for each of the twenty classes of parasites, with a recognition rate of 100%. The measure of root mean square error decreases with the number of epochs; subsequently, that error vanishes while training and evaluating the system, and confirms the high recognition rate achieved by the proposed classifier. Conclusions In this paper, we proposed a neuro-fuzzy system that classifies human intestinal parasites (Protozoa and Helminths) with high accuracy, independent to their stage of development (egg, cyst, or trophozoite). A satisfactory classification for twenty classes of parasites was successfully achieved, with possible extensions in future work.