Obtención de características de subtipos de leucemia en imágenes digitales de células sanguineas para su clasificación
In spite of the recent advances in hematological techniques such as flux cytometry (with immunophenotype), and DNA analysis; morphological analysis of bone marrow smears (even of peripheral blood) are still the starting point to detect patients that suffer of blood disorders. This is why the identif...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Estado: | Versión aceptada para publicación |
| Fecha de publicación: | 2008 |
| País: | México |
| Institución: | Instituto Nacional de Astrofísica, Óptica y Electrónica |
| Repositorio: | Repositorio Institucional del INAOE |
| Idioma: | español |
| OAI Identifier: | oai:inaoe.repositorioinstitucional.mx:1009/435 |
| Acceso en línea: | http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/435 |
| Access Level: | acceso abierto |
| Palabra clave: | info:eu-repo/classification/Segmentación de imagen/Image segmentation info:eu-repo/classification/Clasificación/Classification info:eu-repo/classification/Medidas de las características/Characteristics measurements info:eu-repo/classification/cti/7 info:eu-repo/classification/cti/33 info:eu-repo/classification/cti/3314 |
| Sumario: | In spite of the recent advances in hematological techniques such as flux cytometry (with immunophenotype), and DNA analysis; morphological analysis of bone marrow smears (even of peripheral blood) are still the starting point to detect patients that suffer of blood disorders. This is why the identification of acute Leukemia subtypes from blood cells is an important task due to its use in clinical diagnosis. The classification of these leukemia subtypes from digital images of blood cells helps the physician to prescribe a suitable treatment to the patient. This work presents a method to generate descriptive characteristics for the identification and classification of acute Leukemia subtypes from digital images of blood cells. The first part of this work consists of a pre-processing phase to segment the image by color to then detect the boundaries of the cells using the three bands of the image: R, G, and B. In the second phase we use the preprocessed images to obtain their descriptive characteristics: texture, geometric, statistical, and their eigenvalues (ACPs) with 80% of variability. These characteristics were used as input attributes to perform the data mining process (using different classifiers) to recognize five different leukemia subtypes. Since our leukemia database presented the class imbalance problem (because of the different proportion of cases of each leukemia subtype), we applied over-sampling techniques to reduce its impact. The evaluation of the results was done by the domain experts Dr. José E. Alonso Chávez, Dr. Rubén Lobato Tolama, and the chemistry Laura O. Olvera Oropeza from the “Instituto Mexicano del Seguro Social” (IMSS) San Jose in Puebla. We also performed a quantitative evaluation using the cross validation technique. Our results for each leukemia subtype were around 85% of accuracy. As result we obtained a set of descriptive characteristics to describe acute leukemia subtypes that allowed us to classify them with a global precision of 88 %. We achieved an accuracy of 85% for subtypes L1 and L2 and 91% for subtypes M2, M3, and M5. With these results we outperformed the average classification accuracy obtained by domain experts, whose error ranges from 20% to 30% [40]. |
|---|