Development of deep learning-based precision tools for Plasmodium and Cryptosporidium parasites analysis from microscopic images ; Développement d'outils de précision basés sur l'apprentissage profond pour l'analyse des parasites Plasmodium et Cryptosporidium à partir d'images microscopiques

In this thesis, we have proposed two main contributions related to the Malaria parasite and Cryptosporidium parasite analysis from microscopic images using deep learning techniques.More specifically, in the first contribution, we have proposed a framework for diagnosing Malaria infection in humans using microscopic images of thin blood smears. Compared to the state-of-the-art studies, it is rather based on a straightforward segmentation and classification approaches, permitting the analysis of the parasite itself instead of the cellcontaining it. In this sense, the framework permits to directly segment the Malaria parasite and to distinguish its species among four major classes: P. Falciparum, P. Ovale, P. Malaria and P. Vivax. We demonstrate the efficiency of our framework and notably its potential of generalization over interclass data by exploiting several public datasets. Moreover, we show that our proposed data augmentation technique named Local Parasite Texture Scanning (LPTS) further improves the accuracy of our classification model.In the second contribution, we have proposed a framework for diagnosing Cryptosporidium infection in dairy cows using fluorescence microscopic images. To this end, we have proposed an original parasite segmentation methodology based on a coarse-to-fine approach which achieves high accuracy on our generated dataset of Cryptosporidium and permits to outperform segmentation methods from the state-of-the-art. We have also proposed a classifier with a high discriminatory power that is used to efficiently distinguish the life stages of the parasites among 4 asexual stages: oocyst, trophozoite, meront, and free form. We show through an experimental study that our classifier achieves high accuracy by analyzing only the parasite itself and without the need of additional information related to the size and the number of nuclei which are required by the biologist to establish the classification. ; Dans cette thèse, nous avons proposé deux contributions principales liées à l'analyse par ...