RESUMEN

El cáncer de cérvix uterino representa una de las mayores amenazas de muerte por cáncer entre las mujeres. Con el avance continuo en la medicina y la tecnología, las muertes por esta enfermedad han disminuido significativamente. Las investigaciones referentes a este tema han podido determinar síntomas claves que permiten detectar a tiempo esta enfermedad para darle un tratamiento oportuno. La citología convencional es una de las técnicas más utilizadas, siendo ampliamente aceptada, de bajo costo, y con mecanismos de control. Con el objetivo de aliviar la carga de trabajo a los especialistas, algunos investigadores han propuesto el desarrollo de herramientas de visión computacional para detectar y clasificar las transformaciones en las células de la región del cuello uterino. La presente investigación tiene como objetivo proveer a los investigadores de una herramienta de clasificación automática, aplicable a las condiciones existentes en los centros médicos y de investigación del país. Esta herramienta debe ser capaz de clasificar las células del cuello del útero, basándose solamente en las características extraídas de la región del núcleo y sin utilizar las características del citoplasma, de manera que se reduzca la tasa de falsos negativos en la prueba de Papanicolaou. A partir del estudio realizado, se obtuvo una herramienta haciendo uso de la técnica k-vecinos más cercanos con la distancia manhattan, el cual mostró un alto desempeño manteniendo valores de AUC superiores al 91 por ciento y llegando hasta un 97.1 por ciento con respecto a los clasificadores SVM y RBF Network, los que también fueron analizados(AU)

Cervix cancer is one of the biggest threats of cancer death among women. With continued advances in medicine and technology, deaths from the disease have fallen significantly. The investigations concerning this issue have determined key symptoms to detect the disease in time to give timely treatment. Conventional cytology is one of the most widely used techniques, being widely accepted, inexpensive, and with control mechanisms. In order to alleviate the workload of specialists, some researchers have proposed the development of computer vision tools to detect and classify the changes in the cells of the cervical region. This research aims to provide a tool for automatic classification, applicable to medical conditions and research centers of the country. This tool should be able to classify the cells of the cervix, based solely on the features extracted from the core region without using the characteristics of the cytoplasm, so that the rate of false negative Pap test is reduced. From the study, a tool is obtained using the k nearest-neighbors manhattan distance technique, which showed a high performance maintaining AUC values greater than 91 percent and reaching 97.1 percent over classifiers SVM and RBF Network, which were also analyzed(AU)

Asunto(s)

RESUMEN

Peripheral T-cell lymphomas are a group of rare neoplasms originating from clonal proliferation of mature post-thymic lymphocytes with different entities having specific biological characteristics and clinical features. As natural killer cells are closely related to T-cells, natural killer-cell lymphomas are also part of the group. The current World Health Organization classification recognizes four categories of T/natural killer-cell lymphomas with respect to their presentation: disseminated (leukemic), nodal, extranodal and cutaneous. Geographic variations in the distribution of these diseases are well documented: nodal subtypes are more frequent in Europe and North America, while extranodal forms, including natural killer-cell lymphomas, occur almost exclusively in Asia and South America. On the whole, T-cell lymphomas are more common in Asia than in western countries, usually affect adults, with a higher tendency in men, and, excluding a few subtypes, usually have an aggressive course and poor prognosis. Apart from anaplastic lymphoma kinase-positive anaplastic large cell lymphoma, that have a good outcome, other nodal and extranodal forms have a 5-year overall survival of about 30%. According to the principal prognostic indexes, the majority of patients are allocated to the unfavorable subset. In the past, the rarity of these diseases prevented progress in the understanding of their biology and improvements in the efficaciousness of therapy. Recently, international projects devoted to these diseases created networks promoting investigations on T-cell lymphomas. These projects are the basis of forthcoming cooperative, large scale trials to detail biologic characteristics of each sub-entity and to possibly individuate targets for new therapies.