Multikernel embedded fusion UNet (MKEF-UNet): a robust deep learning approach for accurate segmentation of Chagas parasites

This paper introduces a novel approach for segmenting Chagas parasites on stained blood smear samples from mice during the acute phase of infection with Trypanosoma cruzi utilizing a U-Net-based deep learning model named multikernel embedded fusion UNet (MKEF-UNet). Our proposed model incorporates DenseNet-121 for feature extraction, a classifier module for predicting parasite information, and a segmentation decoder with multiscale feature fusion to generate precise segmentation results. Notably, the integration of the embedded vector module, multikernel convolutions with dilations, and advanced data augmentation techniques significantly enhance the model’s robustness and generalization capabilities. In extensive experiments on the Chagas dataset, MKEF-UNet achieves state-of-the-art performance, attaining an intersection over union (IoU) of 0.8683 and a Dice score of 0.9189 on the test set. The paper provides a detailed exposition of the model’s architecture, training methodology, and the employed loss functions. This research not only presents a novel segmentation approach but also underscores the model’s superiority in accuracy and robustness through the incorporation of advanced features. By introducing robust components and techniques, MKEFUNet surpasses previous methods, showcasing its potential as a promising solution for the detection and segmentation of Chagas parasites in real-world scenarios.