Utilidad diagnóstica en el glaucoma del análisis de las fibras retinianas mediante polarimetría láser asociado a la autoperimetría y a sistemas de inteligencia artificial (redes neuronales)

Resumen del Libro

Owing to the great importance of Glaucoma and because it is one of the main causes of blindness in the world, we are working to contribute a new help to obtain a diagnosis as early as possible, an adequate statistics and an early treatment for Glaucoma, so that we could improve the quality of life of persons who suffer this problem. The persons who have been included in the population sample were voluntarily included and they came from ophthalmic routine exams made in their practice, they signed the consent to be included in the protocol and they declared not taking any medication which could affect the results, in the same ways, they also declared not being regular drinkers or smokers. 106 eyes of 53 patients have been studied (31 women and 22 men) each eye was considered individually. The eyes are classified in different stages of Glaucoma, defined by an expert ophthalmologist, taking into account the definition proposed by Caprioli (1992), in which eyes are defined from group 0 (normal eyes) up to group 5 (terminal Glaucoma) from obtained data from the medical history (qualitative variable), from contributions of the visual area study by autoperimetry and from the study of nerve fiber layer of the retina by polarimetry laser (quantitative variable). The Artificial Intelligence is applicated, it is a neuronal network system which provide a degree of objectivity as much as possible when establishing a clinical diagnosis of Glaucoma. When we have finished the design and training of the system, the classification of the Artificial Intelligence will be able to assign each patient their corresponding group without the especialist taking part with a high index of specificity and sensitivity. Among the conclusions we underlined that the assessment of the visual memory by the program seven of the autoperimetry kinetic/static Dicon TKS 4000 is a fat and reliable test to differentiate healthy from pathological persons in advanced stages of Glaucoma without providing algorithms which can use to quantify the depth of the zone and their relation with incipient alterations. The functional capacity and measures of perimetric sensitivity quadrant are suitable parameters to be considered as input variables in neuronal network design in this study as evaluation of visual field defects by autoperimetry objectified. The analysis of retinal never fiber layer defects by scanning laser polarimetry using the GDX analyzer NFA II version, suppose an indirect quantitative method which objectifies peripapillary level changes. Among the index polarimetrics which we have obtained, we can use them as input variables of the neuronal net as follows: desviations from normal rates in each quadrant, the number of fibers and average thickness. The neuronal network multilayer perceptron of backpropagation shape in this study has the ideal caracteristics with 16 input neurons 30 hidden layer nodes and 1 neuron in the output layer requiring 8000 cycles with training features to achivie the minimum mistakes of the neural net. The neural network design achivies an efficiency of 100% in the discrimination and assignment to each stage of Glaucoma established by the specialist expert. Neuronal network integration proposed in this study, in the functional exploratory test and structural of affectation in the Glaucoma will allow to implement the ability to diagnosis and classification without high requeriments, there is no doubt that Neuronal network integration is an useful tool to interpret the results in terms of glaucomatous affectation, prognosis and therapeutic guidance.