| |
| Acadêmico(a): Juliano Mueloschat |
| Título: VONCELL: UM PROTÓTIPO PARA RECONHECIMENTO DE CÉLULAS DO SISTEMA IMUNE DO TIPO LINFÓCITO E NEUTRÓFILO |
| |
| Abstract: |
| This work presents a prototype that performs the recognition of typical leukocytes of the lymphocyte and neutrophil type through images captured from an optical microscope. The flow of the prototype was divided into two phases: field enhancement and cell discovery, and isolation and extraction of characteristics of the nucleus and cytoplasm to infer cell classification. The first phase is the field enhancement step, which uses the input image, converting it from RGB to the HSV color system. The V-channel is used to delineate the region of the field, whereas the S-channel presents the enhancement of the nuclei of the cells. From this, the S channel undergoes several noise elimination steps to separate the core from the remainder of the image. After extracting the bottom of the image, the nuclei undergo regional growth, thus creating the region of the cells. Cell regions are clippings of the input image. The second phase separates the nucleus and cytoplasm. They are used in the extraction and calculation of the morphological characteristics of the cell. These characteristics are used to infer the classification of the cell through an MLP neural network. The neural network used 22 cells, 7 lymphocytes and 15 neutrophils, to perform the training, on a base of 40 images with 89 cells. The results show that the prototype obtained a percentage of 94.7% accuracy in the total cell count. While in the differential count, it obtained 95% accuracy in lymphocyte classification and 94.4% in neutrophil. |
| |
|
