Mostrar el registro sencillo del ítem

dc.contributor.authorGonzález López, Antonio Francisco 
dc.date.accessioned2016-01-19T09:15:30Z
dc.date.available2016-01-19T09:15:30Z
dc.date.issued2015-07
dc.description.abstract[SPA] En esta Tesis se ha estudiado la mejora de la calidad de la imagen portal, siendo el objetivo principal la reducción de ruido. La imagen portal se estudia en el dominio wavelet para comprobar en qué medida se asemejan sus propiedades estadísticas a las encontradas en imagen natural. El mismo estudio llevado a cabo sobre la imagen portal se realiza sobre el ruido de los sistemas de imagen portal. Posteriormente se prueban diversos métodos de reducción de ruido sobre imágenes portales. De entre los métodos de reducción de ruido, los que operan en el dominio wavelet son la base del estudio presentado en esta Tesis. Además, dos métodos como el filtro de Wiener y el filtro de medias no locales NLM, que operan en el dominio de la imagen, sirven como referencia con la que comparar a los filtros wavelet. Se encuentra que el ruido de los sistemas de imagen portal tiene una distribución estadística de tipo gausiano, y que su espectro es aproximadamente plano en el caso de uno de los detectores estudiado. Además, el valor medio del ruido no depende del valor de la señal, mientras que la varianza del ruido es aproximadamente proporcional al valor de la misma. La distribución estadística del ruido en el dominio wavelet es gausiana y no se observan las dependencias estadísticas entre los coeficientes wavelet observadas en las imágenes del entorno humano. En este sentido, las funciones densidad de probabilidad conjuntas y condicionadas intra-banda e inter-banda son similares a las obtenidas para el ruido blanco y gausiano. El espectro promedio de un conjunto de imágenes portales tiene una dependencia con la frecuencia espacial f según f-a, con a=3.7. Este comportamiento dista del encontrado para las imágenes del entorno humano, en el que la dependencia era del tipo f-2. Esto supone que, para las imágenes portales, la distribución equitativa de la potencia entre las décadas del espectro no se cumple. Por tanto, la pérdida de contraste en las estructuras al disminuir su tamaño es muy importante. Las funciones densidad de probabilidad marginales, para los coeficientes wavelet de las imágenes portales, se asemejan a las obtenidas para las imágenes del entorno humano. Además, las relaciones estadísticas entre los coeficientes wavelet, en una misma banda o en diferentes bandas, también son similares a las encontradas en las imágenes del entorno humano. En cuanto a la reducción de ruido, todos los métodos wavelet analizados obtienen resultados sensiblemente mejores que el filtro de Wiener adaptativo. De entre los filtros que operan en el dominio wavelet, el método bayesiano BLS-GSM obtiene los mejores valores de PSNR en la mayoría de los casos estudiados. Muy cerca de él se sitúan los filtros de umbralización no diezmada y el filtro SURE, de minimización del estimador insesgado de riesgo de Stein. Los tiempos de cálculo varían considerablemente entre los algoritmos estudiados. Para el BLS-GSM los tiempos oscilan entre los 10s y los 20s. Para el resto de algoritmos wavelet el tiempo de cálculo está condicionado por el tipo de transformación que utilizan, diezmada o no diezmada. Así, el algoritmo de umbralización diezmada emplea unas pocas décimas de segundo, mientras que el algoritmo de umbralización no diezmada emplea unos tiempos aproximadamente 10 veces mayores. Por último, el algoritmo de minimización del SURE, que emplea transformaciones diezmadas, realiza los cálculos en menos de 1s. Otros temas que son objeto de estudio en la presente Tesis son la resolución espacial, el procesamiento wavelet y el procesamiento de la imagen en la dosimetría con película en radioterapia. Así, se estudia la resolución espacial de los sistemas de imagen portal; se presenta un nuevo método para la determinación de la resolución espacial de los equipos de imagen de radiología digital: a partir de las imágenes de un patrón de barras en estrella, se desarrolla un algoritmo de agrupación, promediado y sobremuestreo angular para el cálculo de la función de transferencia de modulación. Los resultados obtenidos se comparan con los del método de referencia, recomendado por la CEI, y se comprueba su mayor inmunidad al ruido; se estudia el cálculo del espectro de potencia en el dominio wavelet; se investiga la reducción de la incertidumbre en la dosimetría con película radiocrómica; se presenta un método para la dosimetría de los campos pequeños de radiación con película radiocrómica; se determina la resolución de señal óptima, en función del nivel de ruido, en el proceso de digitalización de películas y se establece el rango de densidades ópticas útil en función de la incertidumbre en un sistema densitométrico basado en un escáner de sobremesa.es_ES
dc.description.abstract[ENG] Portal images are used in radiotherapy for the verification of patient positioning. The distinguishing feature of this image type lies in its formation process: the same beam used for patient treatment is used for image formation. This feature makes the portal image a very important tool in the current and future radiation therapy, especially in applications such as gating or tracking, key in the real time monitoring of radiotherapy treatments. The high energy of the photons used in radiotherapy strongly limits the quality of portal images. The interaction probabilities of a photon by means of Compton process, radiationmatter interaction prevalent in radiotherapy, with soft tissue and bone are not as different as they are for the photoelectric effect (predominant in radio-diagnostic). This results in a very low contrast between biological tissues. In addition, high energy photons limit the image resolution and detection efficiency of imaging devices. This last effect gives rise to a low signal to noise ratio. The low quality of portal images, due to the use of high-energy photon beams, and their usefulness in radiotherapy treatments, justify the efforts aimed to improve it. This Thesis studies the enhancement of these images, in particular denoising of portal images. Processing natural images in the wavelet domain has achieved great results. Image compression, noise reduction and image registration are examples of applications that have demonstrated the usefulness of wavelets in image processing. This great performance is explained taking into account the statistical characteristics of the image in the wavelet domain, particularly the shape of marginal probability distributions, and the statistical dependencies between the wavelet coefficients. In this Thesis portal images are studied in the wavelet domain to check to what extent their statistical properties resemble those found in natural images. The same study conducted on portal images is performed on the noise of portal imaging systems, and the results are compared with those of a white gaussian noise. Later, various denoising methods are applied to noisy portal images. Then, the results obtained with these methods are analyzed on the basis of the statistical characteristics previously studied. Among the denoising methods studied, those operating in the wavelet domain are the basis of the study presented in this Thesis. In addition, two methods such as the Wiener filter and the non local means filter NLM, operating in the image domain, are used as a reference to compare the wavelet filters. Other topics studied in this Thesis are spatial resolution, wavelet processing and image processing in film dosimetry in radiotherapy. In this regard, the spatial resolution of portal imaging systems is studied; a new method for determining the spatial resolution of the imaging equipments in digital radiology is presented; the calculation of the power spectrum in the wavelet domain is studied; reducing uncertainty in film dosimetry is investigated; a method for the dosimetry of small radiation fields with radiochromic film is presented; the optimal signal resolution is determined, as a function of the noise level and the quantization step, in the digitization process of films and the useful optical density range is set, as a function of the required uncertainty level, for a densitometric system. The study of the statistical distribution of portal images, and noise from portal imaging systems is performed on sets of images obtained with two systems based on different technologies. The anatomical portal images were images obtained during the treatment of patients, while noise images were obtained from images of a uniform methacrylate phantom. The imaging systems used were aS500 (Varian Medical Systems) and iView (Elekta), integrated in a linear accelerator 2100 Clinac DHX (Varian) and in a Precise linear accelerator (Elekta) respectively. Power spectra for portal images and noise, statistical dependencies between image and noise as well as marginal, joint and conditional statistical distributions in the wavelet domain are the statistical characteristics studied for images and noise. MATLAB (Mathworks Inc.) was used for analyzing the statistics of the image ensembles. It was also used for implementing the denoising methods and for the assessment of the denoising results. MATLAB was also used for generating synthetic images used in the study of the spatial resolution of imaging systems, for the implementation of the algorithms for calculating the modulation transfer function, for studying the spectrum in the wavelet domain and for reading and processing the radiochromic films. PinPoint and Semiflex (PTW Freiburg) ionization chambers and Radiochromic film EBT2 (International Specialty Products) were used for the dosimetry of the radiation fields. The measurements were carried out on methacrylate phantoms, and the films were digitized in a ScanMaker 9800XL (Microtek) flatbed scanner. The transversal profiles of small fields measured with radiochromic films were verified with a Monte Carlo simulation. The Monte Carlo code used was PENELOPE (Penetration and Energy Loss of Electrons and positrons). It is found that noise in portal imaging systems has a statistical distribution of Gaussian type, and its spectrum is approximately flat in the case of one of the detectors studied. Furthermore, the average noise value does not depend on the value of the signal. However, the noise variance is roughly proportional to the value of the signal. The statistical distribution of the noise is also Gaussian in the wavelet domain and the statistical dependencies between wavelet coefficients observed in natural images are not observed for noise. In this regard, the joint and conditional probability density functions, intra-band and inter-band, are similar to those obtained for white Gaussian noise. When comparing the two imaging systems analyzed, the most advanced technologically (aS500 with an amorphous silicon detector) obtains a signal-to-noise ratio greater than the CCD detector for all spectrum frequencies where comparison is possible. The averaged spectrum for the ensemble of portal images has a dependence on the spatial frequency f as f^-a, with a=3.7. This behavior is far from that found for natural images, in which the spectrum has a dependence as f^−2. This means that, for portal images, the energy is not equally distributed between the decades of the spectrum. Therefore, the loss of contrast in the structures as their size reduces is very important. On the other hand, the marginal probability density functions, for wavelet coefficients of portal images, resemble those obtained for natural images. In addition, the statistical relationships between the wavelet coefficients are also similar to those found in natural images. Regarding noise reduction, all the analyzed wavelet methods obtain significantly better results than the adaptive Wiener filter. The non local mean filter NLM, which operates in the domain of the image, obtained denoising results comparable to the best results obtained with filters that operate in the wavelet domain. Among the filters that operate in the wavelet domain, the Bayesian method BLS-GSM obtains the best PSNR values in most of the cases studied. Very close to it we found filters as the undecimated thresholding and the SURE filter. Calculation time, fundamental in radiotherapy, vary considerably between the algorithms studied. For the NLM algorithm the calculation time exceeds 1 min in most cases. Calculation time for BLS-GSM ranges from 10s to 20s. For the rest of the wavelet algorithms the calculation time is determined by the type of transformation used, decimated or undecimated. Thus, the decimated thresholding algorithm employs a few tenths of a second, while the undecimated thresholding algorithm employs a few seconds. Finally, the SURE minimization algorithm, which uses decimated transformations, performs computations in less than 1s. In summary, wavelet processing demonstrates to be efficient in improving the quality of portal images by means of denoising. Considering the quality of the restored image and the calculation time, methods operating in the wavelet domain outperform the results of the methods that operate in the image domain. Other topics covered in this Thesis, closely related to the main purpose of it, are: A method for determining the modulation transfer function in diagnostic imaging equipment. From the image of a star bar pattern an oversampling, bining and averaging algorithm is developed. The outcome of the algorithm is a set of circular scanning functions, from which it is possible to determine the MTF of the imaging system. The results obtained following the presented method are compared with the reference method, recommended by the IEC, and show a better immunity against noise. A study on the determination of the spectrum of an image in the wavelet domain. The study links the errors in the calculated spectrum with the support of the wavelets used, their symmetry, and the methods followed for the extension of the image bands. The distortions are minimal for very symmetrical wavelets of small support and high number of vanishing moments (as sym8 or sym16). Also, the spectra are more accurately calculated when symmetrical band extensions are carried out, and the extended parts of the image bands are excluded from the calculations. A method for radiochromic film dosimetry. The method significantly reduces one of the most important problems in this dosimetry: the spatial inhomogeneity of the film and digitizer response. The standard deviation of the measurements is reduced from 4.4 % to 0.7 %. This dosimetric method is then adapted to be used in the dosimetry of small radiation fields. The dose is determined in the central axis, along the transversal profiles and in 2D dose distributions for field sizes down to 0.5 × 0.5cm2. The central axis measurements were validated with central axis measurements using ionization chambers in larger fields. The dose measurements along the transverse profiles were validated with Monte Carlo simulations. An investigation on the accuracy of film dosimetry using a digitizer, later used in the study of the resolution of portal imaging systems. Two main results are found. First, the optimal signal resolution of the digitizer for a given analog noise: It is found that, if the quotient s_an/D, where D is the quantization step and s_an is the standard deviation of noise, is less than 0.6 then the quantization noise dominantes the system noise. On the other hand, if this ratio increases above 0.6 no sensitive improvement of the total noise is achieved. Second, the optical density range of a film-digitizer system, for a given uncertainty, is mathematically expressed as a funcion of the uncertainty. The efficiency of wavelet processing in portal imaging denoising suggests to investigate other improvements in image quality, derived from processing the image in the wavelet domain. Portal imaging restoration, aimed to improve the perception of the image detail, portal imaging registration with reference images (simulation images) and denoising in video with portal imaging are possible lines of future research. Further studying noise, spatial resolution of medical imaging systems and dosimetry of small radiation fields are other open issues suggesting future lines of research.es_ES
dc.formatapplication/pdfes_ES
dc.language.isospaes_ES
dc.publisherAntonio Francisco González Lópezes_ES
dc.rightsAtribución-NoComercial-SinDerivadas 3.0 España*
dc.rights.urihttp://creativecommons.org/licenses/by-nc-nd/3.0/es/*
dc.title.alternativeDenoising portal images by means of wavelet techniqueses_ES
dc.titleReducción de ruido en Imagen Portal mediante técnicas de Waveletes_ES
dc.typeinfo:eu-repo/semantics/doctoralThesises_ES
dc.contributor.advisorMorales Sánchez, Juan 
dc.date.submitted2015-11-27
dc.subjectImagen portales_ES
dc.subjectRuidoes_ES
dc.subjectSistemas de imagenes_ES
dc.subjectTecnología médicaes_ES
dc.subjectFísica médicaes_ES
dc.subjectDominio waveletes_ES
dc.subjectPortal imaginges_ES
dc.subjectImage statisticses_ES
dc.subjectWavelet processinges_ES
dc.subjectDenoisinges_ES
dc.identifier.urihttp://hdl.handle.net/10317/5204
dc.contributor.departmentTecnologías de la Información y las Comunicacioneses_ES
dc.identifier.doi10.31428/10317/5204
dc.rights.accessRightsinfo:eu-repo/semantics/openAccesses
dc.description.universityUniversidad Politécnica de Cartagenaes_ES
dc.subject.unesco22 Físicaes_ES
dc.subject.unesco32 Ciencias Médicases_ES
dc.subject.unesco3314 Tecnología Médicaes_ES
dc.description.programadoctoradoPrograma de doctorado en Tecnologías de la Información y las Comunicacioneses_ES


Ficheros en el ítem

untranslated

Este ítem aparece en la(s) siguiente(s) colección(ones)

Mostrar el registro sencillo del ítem

Atribución-NoComercial-SinDerivadas 3.0 España
Excepto si se señala otra cosa, la licencia del ítem se describe como Atribución-NoComercial-SinDerivadas 3.0 España