doctoralThesis
Técnica para segmentação automática de imagens microscópicas de componentes sanguíneos e classificação diferencial de leucócitos baseada em lógica fuzzy
Fecha
2014-12-26Registro en:
VALE, Alessandra Mendes Pacheco Guerra. Técnica para segmentação automática de imagens microscópicas de componentes sanguíneos e classificação diferencial de leucócitos baseada em lógica fuzzy. 2014. 100f. Tese (Doutorado em Engenharia Elétrica e de Computação) - Centro de Tecnologia, Universidade Federal do Rio Grande do Norte, Natal, 2014.
Autor
Vale, Alessandra Mendes Pacheco Guerra
Resumen
Automatic detection of blood components is an important topic in the field of
hematology. The segmentation is an important stage because it allows components to be
grouped into common areas and processed separately and leukocyte differential classification
enables them to be analyzed separately. With the auto-segmentation and differential
classification, this work is contributing to the analysis process of blood components by
providing tools that reduce the manual labor and increasing its accuracy and efficiency.
Using techniques of digital image processing associated with a generic and automatic fuzzy
approach, this work proposes two Fuzzy Inference Systems, defined as I and II, for autosegmentation
of blood components and leukocyte differential classification, respectively, in
microscopic images smears. Using the Fuzzy Inference System I, the proposed technique
performs the segmentation of the image in four regions: the leukocyte’s nucleus and
cytoplasm, erythrocyte and plasma area and using the Fuzzy Inference System II and the
segmented leukocyte (nucleus and cytoplasm) classify them differentially in five types:
basophils, eosinophils, lymphocytes, monocytes and neutrophils. Were used for testing 530
images containing microscopic samples of blood smears with different methods. The images
were processed and its accuracy indices and Gold Standards were calculated and compared
with the manual results and other results found at literature for the same problems.
Regarding segmentation, a technique developed showed percentages of accuracy of 97.31%
for leukocytes, 95.39% to erythrocytes and 95.06% for blood plasma. As for the differential
classification, the percentage varied between 92.98% and 98.39% for the different leukocyte
types. In addition to promoting auto-segmentation and differential classification, the
proposed technique also contributes to the definition of new descriptors and the construction
of an image database using various processes hematological staining