17April2026
09:00 Master's Defense IC room 85
Topic on
A Study on Lightweight Networks for the Classification of Enteroparasites
Student
Maria Angélica Krüger Miranda
Advisor / Teacher
Alexandre Xavier Falcao
Brief summary
Enteroparasitoses represent a serious global public health problem, disproportionately affecting populations in regions with poor sanitation. The gold standard laboratory diagnosis relies on manual optical microscopy, a slow, tiring process dependent on the technician's expertise. Although automation via deep learning has shown potential to aid in this diagnosis, traditional convolutional neural networks have critical limitations for large-scale deployment: they require high computational power and depend on large volumes of annotated data, which are scarce for some parasite species. This dissertation describes the use of Feature Learning from Image Markers (FLIM) for the classification of enteroparasites, a method that builds feature extractors from image markers without the need for backpropagation. Unlike the original technique, which depends on user intervention for marker selection, this approach automates the process using the Dynamic Iterative Spanning Forest (DISF) algorithm. Starting from the segmentation of the region of interest into superpixels, a set of characteristic points is constructed that serve as the basis for assembling the convolutional filters of the network. The method was evaluated on three datasets (helminth eggs, helminth larvae, and protozoan cysts) obtained via TF-Test technique and DAPI system. The experimental protocol includes a comprehensive comparative analysis, involving state-of-the-art lightweight network architectures, as well as networks trained from scratch and other variations, under a low-data learning scenario. The results demonstrate that the approach is competitive, maintaining stable accuracy, F1-Score, and Kappa indices even with reduced training sets, presenting a lower number of parameters than the evaluated reference architectures.
Examination Board
Headlines:
| Alexandre Xavier Falcão | IC / UNICAMP |
| Fabio Augusto Faria | IST/ULisboa |
| Marcelo da Silva Reis | IC / UNICAMP |
Substitutes:
| Edson Borin | IC / UNICAMP |
| Fátima de Lourdes dos Santos Nunes Marques | EACH / USP |