Publication: Diseño e implementación en FPGA de un filtro Kalman Unscented para aplicaciones biomédicas
Loading...
Official URL
Full text at PDC
Publication Date
2017
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
Actualmente, la vida de los pacientes con Diabetes Mellitus está muy controlada. Su enfermedad no les permite mantener un nivel estándar de glucosa en sangre y necesitan de insulina externa para bajarla, ya que este tipo de diabetes suele generar altos niveles de glucosa en sangre o hiperglucemia, que es peligroso para su vida. Otra causa de muerte más peligrosa todavía es el caso contrario, la falta de glucosa en sangre o hipoglucemia, que se puede dar por no llevar una buena dieta o por pasarse en la dosis de insulina. Para mejorar la calidad de vida de los diabéticos se han desarrollado durante años Monitores Continuos de Glucosa (MCG). Estos miden la Glucosa Intersticial (GI) entre las capas de la piel, derivada de la Glucosa en Sangre (GS). La GI está relacionada con la GS, pero no son el mismo valor. Debido al retardo provocado por todas las capas de piel, hasta llegar a la zona intersticial la GS se va difundiendo y bajando. Además, la medición de la GI es ruidosa y degenerada por la degradación del sensor llamada ganancia del sensor.
Un gran cambio en los MCG fue el inicio de la aplicación de Filtros Kalman para estimar el valor de la GS. La capacidad de este filtro reside en que, con medidas indirectas de una variable que se quiera estimar, es capaz de pronosticar el estado oculto, que es el valor de la variable que nos interesa.
El objetivo de este Trabajo de Fin de Grado es usar un Filtro Kalman Unscented para la estimación de la GS (estado oculto) a través de las medidas de un sensor de glucosa situado debajo de la piel del paciente y que mide la GI (medida rápida, cada cinco minutos). Además se dispondrá de otra medida muy fiable, ya que se medirá la glucosa en sangre directamente mediante un pinchazo en el dedo, para la calibración y corrección del error del filtro (medida lenta, cada ocho horas). El filtro está implementado en una FPGA por sus características de robustez, seguridad y velocidad de cómputo. El resultado del uso del filtro es un sistema capaz de aproximar la GS y la ganancia del sensor a partir de una medida ruidosa fuertemente no lineal.
Actually, Diabetes Mellitus patient's live is so difficult. This disease don't able themselves to manage an standard blood glucose level and need extern source insulin to decrease it because this kind of diabetes uses to raise blood glucose level or hyperglycemia, dangerous for them. Another more dangerous cause of death is the opposite, the low level blood glucose or hypoglycemia, that could be present because of a bad diet or take so much insulin than the normal dose by an accident. To improve the diabetics quality of live Continous Glucose Monitors (CGM) has been developed for years. This devices sample Interstice Glucose (IG) between skin layers, that came from the Blood Glucose (BG). The IG is related to BG, but isn't the same values. Because of all skin layers delay until came into interstice area the BG spreads himself and decrease. In addition, the IG sampling is noisy and degenerate, because of sensor degradation called sensor gain. A grate CGM change was the Kalman Filtering using to BG estimation. The feature of this filter besides in the ability of variable hidden state estimation, that it's what we want, using indirect variable sampling. The goal of this final year dissertation is to use an Unscented Kalman Filter to BG estimation (our hidden state) using IG under skin glucose sensor sampling (fast measure, five minutes frequency). In addition we are able to use an BG sample by an fingerstick device, a very reliable sample, used to filter calibration and sintonization (slow measure, eight hour frequency). The filter is implemented on a FPGA because of it's good features of strenght, security and compute speed. The result is a device that can estimate BG and sensor gain using hard non-lineal measure.
Actually, Diabetes Mellitus patient's live is so difficult. This disease don't able themselves to manage an standard blood glucose level and need extern source insulin to decrease it because this kind of diabetes uses to raise blood glucose level or hyperglycemia, dangerous for them. Another more dangerous cause of death is the opposite, the low level blood glucose or hypoglycemia, that could be present because of a bad diet or take so much insulin than the normal dose by an accident. To improve the diabetics quality of live Continous Glucose Monitors (CGM) has been developed for years. This devices sample Interstice Glucose (IG) between skin layers, that came from the Blood Glucose (BG). The IG is related to BG, but isn't the same values. Because of all skin layers delay until came into interstice area the BG spreads himself and decrease. In addition, the IG sampling is noisy and degenerate, because of sensor degradation called sensor gain. A grate CGM change was the Kalman Filtering using to BG estimation. The feature of this filter besides in the ability of variable hidden state estimation, that it's what we want, using indirect variable sampling. The goal of this final year dissertation is to use an Unscented Kalman Filter to BG estimation (our hidden state) using IG under skin glucose sensor sampling (fast measure, five minutes frequency). In addition we are able to use an BG sample by an fingerstick device, a very reliable sample, used to filter calibration and sintonization (slow measure, eight hour frequency). The filter is implemented on a FPGA because of it's good features of strenght, security and compute speed. The result is a device that can estimate BG and sensor gain using hard non-lineal measure.
Description
Trabajo de Fin de Grado en Ingeniería de Computadores (Universidad Complutense, Facultad de Informática, curso 2016/2017)