J. S. Sahambi

Wavelet domain non-linear filtering for MRI denoising.

Feature-preserved denoising is of great interest in medical image processing. This article presents a wavelet-based bilateral filtering scheme for noise reduction in magnetic resonance images. Undecimated wavelet transform is employed to provide effective representation of the noisy coefficients. Bilateral filtering of the approximate coefficients improves the denoising efficiency and effectively preserves the edge features. Denoising is done in the square magnitude domain, where the noise tends to be signal independent and is additive. The proposed method has been adapted specifically to Rician noise. The visual and the diagnostic quality of the denoised image is well preserved. The quantitative and the qualitative measures used as the quality metrics demonstrate the ability of the proposed method for noise suppression.

LV Contour Extraction from Cardiac MR Images Using Random Walks Approach

Heart failures are of increasing importance due to increasing life expectation. For clinical diagnosis parameters for the condition of hearts are needed and can be derived automatically by image processing. Accurate and fast image segmentation algorithms are of paramount importance for a wide range of medical imaging applications. In this paper, we present a method using heat equation with variable threshold technique towards seeds selection in random walk based image segmentation.

Weighting function in Random Walk based left ventricle segmentation

Cardiac Magnetic Resonance (CMR) image segmentation is a crucial step before physicians go for patient diagnoses, related image guided surgery or medical data visualization. Most of the existing algorithms are effective under certain circumstances. On the other hand, Random Walk approach is robust for image segmentation in every condition. Weighting function plays an important role for a successful segmentation in the approach. In this work, an attempt has been made to study the behavior of the weighting function with respect to the intensity distribution in the object to be segmented. In this work, we present a weighting function viz. derivative of Gaussian, that is proved to yield better segmentation results while applying on ischemic CMR images, where objects are obscure. Virtuous results on CMR images describes the potential of the weighting function.

Detection of Left Ventricular Myocardial Contours from Ischemic Cardiac MR Images

Abstract Quantitative evaluation of cardiac function from cardiac magnetic resonance (CMR) images requires the identification of the myocardial walls. This generally requires the clinician to view the image and interactively trace the contours. The myocardial wall of the left ventricle in CMR images, obtained from subjects having serious diseases, is obscure, henceforth, its detection is a tough task. In this paper, an approach to outlining the left ventricular contour is proposed. The utilization of a random walk approach is shown in order to extract the blood pool boundary or endocardium (the inner side of the left ventricular myocardial wall). Inaccurate segmentation is resulted, while applying the approach to ischemic CMR images, because this type of images bear multilabeled blood pool and the manual seed(s) selection in these images introduces variability in the results. In view of this, the paper presents two modifications in the algorithm: (1) automatic seed(s) selection and (2) introduction of Laplacian of difference of Gaussian weighting function. Subsequently, a modified version of an active contour method is implemented to extract the epicardium (the outer side of the left ventricular myocardial wall). This outer contour is achieved by taking the blood pool boundary as its initial contour. Basically, this method is based on active contour without edges. Promising experimental results in CMR images demonstrate the potentials of our approach.

Image denoising using spatial context modeling of wavelet coefficients

The choice of threshold in wavelet based image denoising is very critical. The universal threshold is a global threshold utilized for denoising the wavelet coefficients. An effective approach for the estimation of universal threshold based on spatial context modeling of the wavelet coefficients has been proposed. Spatial contextmodeling involves determination of the correlated pixels within a local neighborhood of the pixel to be denoised. Thus the threshold determination depends on the pixel characteristics and not on the size of the image to be denoised. The spatial context information of the wavelet coefficients are computed using the range filter employed in the formation of bilateral filter. Experiments on several Gaussian noise corrupted images show that the proposed method outperforms other thresholding methods such as VisuShrink, SureShrink and BayesShrink.

Effect of β in random walk approach for LV contour extraction

In todays world, increasing life expectation have made the heart failures of important concern. For clinical diagnosis, parameters for the condition of heart are needed. Accurate and fast image segmentation algorithms are of paramount importance prior to the calculation of these parameters. An automatic method for segmenting the cardiac magnetic resonance (CMR) images is always desired to increase the accuracy. We prefer random walk method due to its noise robustness and unconditional approach over other segmentation algorithms. Performance of the method solely depends on the selection of the free parameter β, which uses to be decided manually. The accuracy of the output significantly depends on this parameter. In this work, we present a method to decide its value automatically enhancing the accuracy of the performance.

LV Contour Extraction Using Difference of Gaussian Weighting Function and Random Walk Approach

Image segmentation is the first step prior to any medical analysis. With the increase in modern disease variety, the images (specially cardiac magnetic resonance (CMR) images) to be segmented are found complex in nature. That might be due to noise, color geometry etc. Random walk method is proved to be good enough to this type of images. Simultaneously, it is robust noise and it does not require any pre-condition to perform. In the present paper we show the importance of weighting function, that is used in the method, on the algorithm output. This paper presents a new approach using difference of Gaussian (DoG) weighting function in the random walk method. We compare the frequently used Gaussian weighting function with DoG and show DoG to be the better one. Finally using DoG weighting function, the random walk method is performed on CMR data for left ventricle contour extraction. The result using DoG weighting function is found to be encouraging than that of Gaussian weighting function. Index Terms—Cardiac magnetic resonance image, Gaussian weighting function, difference of Gaussian weighting function, random walker.

Noise Effect on LV Image Segmentation

According to the basic knowledge of the information theory, noise is known to hinder signal quality, and as the noise level increases the signal detection sensitivity decreases. Noise has a detrimental effect on tasks involving vigilance, memory and divided attention. Its effects vary depending on the nature of the noise (including volume, predictability and perceived control) and the type of task that participants are asked to undertake. Rician noise introduces a bias into MRI measurements that can have a significant impact on the cardic magnetic resonance (CMR) image segmentation. Noting to the variations of the noise with the signal amplitude, this paper discusses the over all effect of noise on segmentation. The eigen values that describe the contrast of the image are shown to have been decreased with the addition of noise. The presence of heavy noise is shown to lead to an under segmentation. Index Terms—Cardiac magnetic resonance image, noise, eigen values, variogram, power spectral density.

A Level Set Method for Cardiac Magnetic Resonance Image Segmentation: An Adaptive Approach

Heart failures are of increasing importance due to increasing life expectation. For clinical diagnosis parameters for the condition of hearts are needed and can be derived automatically by image processing. Accurate and fast image segmentation algorithms are of paramount importance for a wide range of medical imaging applications. Level set algorithms based on narrow band implementation have been among the most widely used segmentation algorithms. The narrow band level set method is a kind of technique that tracks the evolving interface. Its computation domain is set near the zero level set. In this work, we present an adaptive method to extract the left ventricle (LV) irrespective of the intensity variation in heart MR data using a narrow-band level set method. Instead of using the image directly, its scaled down versions are used removing the unnecessary redundancies and extra computations. Also, we suggest an automatic approach for gaussian parameter selection.

Modified random walk approach for left ventricular contour extraction from cardiac MR images

From a proper diagnosis point of view, an accurate and fast image segmentation algorithms are of paramount importance prior to the calculation of the diagnostic parameters. A precise method for segmenting the cardiac magnetic resonance (CMR) images is always desired to increase the accuracy. We prefer random walk method, as it is robust to noise and does not require any pre-knowledge about the input image, on which the segmentation is to be performed. This paper suggests a new approach towards CMR image segmentation using modified random walk approach which improves the over-all performance of the fixed random walk approach. The average number of steps and computational time in this method are much less than that of a fixed random walk with better segmentation results.