A framework for segmentation of filarial worm in thick blood smear images using image processing techniques and machine learning algorithms

Abstract

Lymphatic filariasis (LF), also referred to as elephantiasis, is an infectious disease prevalent in tropical regions, caused by parasitic filarial worms and spread through the bites of mosquitoes. Individuals with LF have difficulty in doing routine tasks, resulting in long-term socioeconomic consequences. Hence, early detection and diagnosis are needed to control the infection. Considering the setbacks during COVID-19 pandemic, the Global Program for Elimination of Lymphatic Filariasis (GPELF) revised the elimination target at 2030. Computer-aided detection and segmentation of microfilariae in microscopic blood smear images is expected to detect the microfilaria more preciously compared to routine microscopic examination, in particular, the weakly-stained smears or coiled microfilaria. In this work, the acquired blood smear images were preprocessed with illumination correction, various filtering, and thresholding methods. It was found that, the 2D Jerman Filter and Renyi entropy-based thresholding resulted in best image quality metrics. Further, five different segmentation algorithms were utilized to segment the filarial worm from the images. It was found that the similarity indices between the ground truth and the images segmented using the firefly algorithm were high with an average Dice, Jaccard, and structural similarity index of 0.9816, 0.9779, and 0.9932, respectively. It is observed that the proposed framework accurately segments the worm without losing its proximal and distal portions, despite the presence of artifacts, and variation in shape and size of the worms due to folding or coiling. This work has significant public health impact since automated segmentation of filarial worms is highly desirable for mass screening of lymphatic filariasis particularly during pre-elimination phase and in low- endemic situation.