Published Papers

Artificial intelligence algorithms are interesting solutions to automate the tedious manual counting of white blood cells by a specialist. Although interesting machine learning algorithms have been proposed for this task, there is a lack in the literature for high-accuracy methods (more than 99%) tested on larger datasets (more than 10 thousand images). This paper presents a segmentation and classification methodology, based on Random Forest and ResNet50, along with a comparison between ResNet models with different numbers of layers.

Photodynamic therapy is a treatment modality that carn be used to treat various types of lesions. To produce cell death, reaching a certain threshold dose of reactive oxygen species (ROS) is required. The estimation of ROS production is of paramount importance to predict the depth of necrosis and to ensure that the volume to be treated receives doses higher than the threshold. In this study, we compared a theoretical model for PDT based on Monte Carlo simulations of light irradiance and rate equations with a rat liver model. At the end of the simulation, necrosis depths and volumes were estimated, as well as the photosensitizer (PS), oxygen, and ROS concentrations at each position of the treated area. From the in vivo study, we obtained the ROS concentration threshold of about 1 mM for Photogem in rat liver. This proposed method can be used for any PS or tissue, including tissues with multiple layers. The proposed method can be used to estimate parameters for any PS or tissue, including layered tissues, as long as their parameters are known. In addition, other protocols can be tested, or compared with the standard ones, providing the bases for analyzing a diverse range of photodynamic treatment scenarios.

This paper presents the development of a hyperspectral imaging system for the classification of H&E-stained histological slides. The system was developed to be coupled to a conventional microscope, with software dedicated to control the instrumentation, to show a colorful live image from an RGB camera, and to acquire the hyperspectral imaging using a liquid crystal tunable filter (LCTF). Hyperspectral images of H&E-stained histological slides undergoing photodynamic therapy were classified with four machine learning algorithms to find damaged tissues (crust). The classification results were presented and show that this technique is promising to classify histological tissue regions.