Denoising and enhancement of medical images

Abstract

Medical imaging captures visual representation of human body’s structural and functional aspects like tissues, bones, blood flow etc. for clinical analysis and medical intervention. Optical Imaging, X-ray, Computer tomography, Magnetic Resonance Imaging (MRI), Ultrasound etc. are common medical imaging techniques used by physicians. Among them ultrasound is the most widely used imaging technique due to its cost effectiveness and human health friendly characteristic. But ultrasound images are inherently corrupted with speckle noise and thus makes physician’s interpretation complex and time-consuming. Therefore, in medical image analysis image denoising has more clinical value since it helps the physicians to reach correct, reliable and speedy diagnosis by mitigating noise from the image. Image denoising also facilitate image segmentation, image fusion, object detection and target recognition. Computer-aided image denoising and image enhancement techniques helps to improve efficiency and accuracy of physician’s interpretation. This research work focused on the development of reliable image denoising and enhancement techniques for echocardiographic images. It aimed to denoise an echocardiographic image without introducing noise distortion and loss of information. Fractional calculus has been used to efficiently mitigate noise of various levels from the echocardiographic image. Also rough set theory and fuzzy logic have been used to draw boundaries between image regions. These concepts helped to handle uncertainty caused by the speckle noise. Three image denoising methods have been proposed in thesis. First proposed denoising methodology performs image denoising in two stages. Stage-1 applies weighted fuzzy mean filter and stage-2 convolves every pixel of the image with a fractional integration filter. Second proposed approach intelligently selects appropriate filter for every image region. Fractional order differintegral filter is proposed in third image denoising methodology. All three proposed denoising schemes not only preserve details in the denoised image but also efficiently reduce noise. Image Enhancement further improves the visual quality of an image. This research also proposes two echocardiographic image enhancement schemes to effectively utilize gradient magnitude and eigenvalue hessian matrix calculations and fractional order derivative concept. Real echocardiographic b-mode images and standard images artificially corrupted with speckle noise have been considered in simulations for this research. Visual and quantitative analysis of simulation results presents significant improvement by the proposed schemes as compared to state-of-the art image denoising and image enhancement techniques.