MATLAB-based Noise Reduction For Fluorescence Molecular Imaging With A GUI

In this thesis,we apply one medical image processing method onto the fluorescence molecular imaging(FMI)technique,used abundantly due to its high sensitivity,low cost and ease of use.With a wide spectrum of light wavelengths and an increased number of fluorescent dyes and proteins,FMI has achieved high resolution scans which enlarged its domain of applications.By combining developed macroscopic techniques and high molecular contrast upon studying the tissues in vivo,this technique turned out to be very powerful.For example,the related fluorescence tomography has the ability to capture images in 3D allowing for improved analysis of tissues.However,and although FMI can noninvasively get the fluorophore distribution in-vivo,the auto-fluorescence of the surrounding organs can affect the fluorescence spectrum of the organ under study,which badly affects the quality of the image.In order to reduce this nuisance,in addition to other forms of background noise,such as charge-coupled device(CCD)currents,one needs a method to reduce these noise factors.Actually,medical images,as an important tool in diagnosis and research,need to be analyzed efficiently in order to provide potential vast information.Thus,any distortion and/or deformation factors affecting the clarity of the image need to be tackled and treated before making full use of the image.Computers are one of the best ways to process images.Converting the 'analog'images into 'digital' ones,allow for digital image processing,which presents a large advantage over traditional means that are affected by errors resulting from limitations related to human operators.Moreover,many not so expensive packages are available already in the market for this objective.We develop in this thesis a method,used previously for one experiment,based on fuzzy C-Means clustering(FCM)suitable for removing the undesired noises,and adjust it to be applicable onto multiple experiments.More precisely,the FCM algorithm partitions a finite collection of elements(pixels)into a less numerous collection of'fuzzy' clusters with respect to some given criterion,where in general one considers neighboring pixels and assign their average value to their centered pixel in a way to decrease edging effects.The method consisted of several stages.In the first smoothing stage,we added many filters in order to remove many kinds of noise,depending on the type of the studied image.This can trivially be generalized to several filters,if needed.We tried the method with the Mean/Median and Wiener filters,where the latter helps to estimate the Fourier transform of the image.In the second stage,we used the Single-level discrete 2-D wavelet transform to single out the main component of the fluorescence signal.By undergoing wavelet transformation,the image gives the coefficients and the detailed coefficients matrices.Upon modifying these matrices by a chosen threshold,and applying the inverse wavelet transform one can reconstruct the original image.In the third FCM stage,we used two inputs,rather than one,from the previous stage in order to refine the clustering process which picks up the main component removing the surrounding background.The stage starts with choosing a number of clusters,then it assigns to each data point random coefficients to represent how much it belongs to the various clusters,and the process repeats itself until it converges.All these stages were implemented in MATLAB environment,which is quite rich with many powerful image processing tools.At the end,we realized one graphical user interface(GUI)in order to assemble all the stages into one easy-to-use step,which is suitable for unexperienced students wishing to test the different filters and their effects.In MATLAB the GUIDE(GUI development environment)provides various tools to design the user interface for custom applications.We checked the validity of the program on medical images obtained from a previous experiment related to a mouse implanted in vivo with a fluorescent bead photographed by a cooled CCD camera.The original image was having clear noise,which was smoothed by applying the filters of the first stage.Applying the wavelet transform,we could extract the main components,while eliminating the background noise with the thresholding process.Finally,we used both the images of the two filters as a double input for the clustering phase,with a single output,and we presented our results showing that we get an acceptable clear image as a result of the de-noising method used in our study.