Detecção do parasita da malária em filmes de gota espessa utilizando redes neurais rasas e técnicas de processamento digital de imagens

Abstract

The Amazon region records about 90% of Brazil's malaria cases and the transmission of this disease is directly related to the region's environmental conditions, mainly within the capital, where the diagnosis is also late due to the shortage of qualified professionals (BRASIL, 2020). This work brings a new perspective on how to detect malaria parasites in thick blood smear sample images by using neural networks and digital image processing techniques. The ultimate goal of this work is to detect the parasite through a shallow neural network that can be embedded in low-memory mobile devices, such as smartphones. Recent work uses complex architectures such as Efficient NET or ROENet, or specialized object detection networks such as YOLO and Faster-RCNN, making it difficult to board the system developed on mobile devices. The techniques presented in this work are based on shallow networks such as simple perceptron, multi-layer perceptron and logistic regressor, all of them associated with methods such as segmentation by Otsu and mathematical morphology. The best results obtained are accuracy: 96.33%, sensitivity: 98.85% and F1-score: 97.57% and outperform previously published literature. However, it is not possible to make a strict comparison between all the work due to the use of different image databases.