Soil transmitted helminth egg detection and classification in fecal smear images using faster region-based convolutional neural network with residual network-50
Donata D. Acula, C. P. L. Buico, G. U. Cruzada, M. K. C. Encelan, C. Y. G. Rivera · IET conference proceedings. · 2023
Intestinal parasitic infections are one of the leading causes of morbidity in areas within tropical and subtropical countries. These infections can cause symptoms such as malnutrition and anemia- factors for growth failure. Soil-Transmitted Helminth, in particular, has been identified as the leading cause of sickness in developing countries. In clinical diagnosis, lengthy inspections could occur despite the guidance of experienced medical laboratory technologists. The researchers aim to circumvent this problem by using Convolutional Neural Networks as systems for the most prevalent species of STH eggs- Ascaris lumbricoides, Trichuris trichiura, and Hookworm eggs. The study evaluated and compared the performance and speed of three CNN architectures, mainly ResNet-50 and Faster R-CNN, with some preprocessing methods derived from Suzuki et al. (2017). The results from the study indicated that Faster R-CNN with ResNet-50 had the highest accuracy with 97.65% out of the three CNN architectures. In terms of efficiency, ResNet-50 was the most efficient with 0.0167 seconds average runtime.