Vikrant Bhateja

A Polynomial filtering model for enhancement of mammogram lesions

This paper presents a preliminary analysis of a class of non-linear filters for enhancement of mammogram lesions. A non-linear filtering approach employing polynomial model of non-linearity is designed by second order truncation of Volterra series expansion. The proposed filter response is a linear combination of Type-0 and Type-II Volterra filters. Type-0 filter provides contrast enhancement, suppressing the ill-effects of background noise. On the other hand, Type-II filter employs edge enhancement. The objective analysis of the proposed filter is carried out by estimating values of quality parameters like CEM and PSNR on mammograms from MIAS and DDSM databases.

Medical image fusion using combination of PCA and wavelet analysis.

Medical image fusion for merging of complementary diagnostic content has been carried out in this paper using Principal Component Analysis (PCA) and Wavelets. The proposed fusion approach involves sub-band decomposition using 2D-Discrete Wavelet Transform (DWT) in order to preserve both spectral and spatial information. Further, PCA is applied on the decomposed coefficients to maximize the spatial resolution. An optimal variant of the daubechies wavelet family has been selected experimentally for better fusion results. Simulation results demonstrate an improvement in visual quality of the fused image in comparison to other state-of-art fusion approaches.

Contrast Improvement of Mammographic Masses Using Adaptive Volterra Filter

Due to ill-performance of X-ray hardware systems, mammographic images are generally noisy with poor radiographic resolution. This leads to improper visualization of lesion details. This paper presents an improved Volterra filter design known as Adaptive Volterra filter for contrast enhancement of mammograms. The operation of the adaptive filter proposed in this work can be classified as Type-0, Type-1 and Type-2 depending upon the nature of background tissues (fatty, fatty-glandular or dense) in the mammogram. This filter is considered as a Taylor series with memory whose truncation to the first non-linear term may lead to a simpler and effective representation. Computer simulations are performed on digital mammograms from MIAS database yielding promising improvement in contrast of the targeted lesion along with reasonable suppression of background in comparison to other enhancement techniques.

An Improved Algorithm for Noise Suppression and Baseline Correction of ECG Signals

Testing and analysis of Electrocardiogram (ECG) signals is one of the major requirements for clinical diagnosis of cardiovascular diseases and deciding future therapies. ECG being a weak non-stationary signal is often interfered by impulse noise as well as baseline drift. This paper presents an improved morphological algorithm for suppression of ailments posed by the above mentioned distortions using non-flat structuring element. Dimensions of the structuring element are optimally selected in a manner to achieve lower distortion rates. Simulation results show significant improvement in baseline correction and noise removal (yielding lower values of error indices and high signal to noise ratios) in comparison to other methods.

A New Morphological Filtering Algorithm for Pre-Processing of Electrocardiographic Signals

Pre-Processing of Electrocardiographic (ECG) signals involves the baseline wander elimination and impulse noise filtering to facilitate automated analysis. In this paper a new morphological filtering algorithm using combinations of flat (two dimensional) structuring elements is proposed for pre-processing of ECG signals. Usage of two dimensional structuring elements, (over single dimension) aids in controlling effective inhibition of noise, leading to reconstruction with minimal signal distortion. Signal to noise ratio (SNR) and Root Mean Squared Error (RMSE) are used as quantitative evaluation measures for optimizing the selection of size of the structuring elements. Experimental results show that the proposed algorithm yields effective pre-processing of ECG signals, thereby eliminating the discussed artifacts.

Design of New Volterra Filter for Mammogram Enhancement

Non-linear filters are generally preferred for image enhancement applications as they provide better filtering results not only by suppressing background noise but also preserving the edges. This paper introduces a new technique for enhancement of digital mammograms using a Volterra filter. The proposed Volterra filter design is obtained by truncation of Volterra series to the first non-linear terms. Truncation of Volterra series leads to a simpler and effective representation without having prior knowledge of higher order statistics. The weight indices of the proposed filter are optimally selected in a manner to provide better enhancement of lesions in the mammograms in comparison to other techniques.

An Evaluation of Edge Detection Algorithms for Mammographic Calcifications

Edge detection is an important module in medical imaging for diagnostic detection and extraction of features. The main limitation of the existing evaluation measures for edge detection algorithms is the requirement of a reference image for comparison. Thus, it becomes difficult to assess the performance of edge detection algorithms in case of mammographic features. This paper presents a new version of reconstruction estimation function for objective evaluation of edge enhanced mammograms containing microcalcifications. It is a non-reference approach helpful in selection of most appropriate algorithm for edge enhancement of microcalcifications and also plays a key role in selecting parameters for performance optimization of these algorithms. Simulations are performed on mammograms from MIAS database with different category of background tissues; the obtained results validate the efficiency of the proposed measure in precise assessment of mammograms (edge-maps) in accordance with the subjectivity of human evaluation.

PCA Based Medical Image Fusion in Ridgelet Domain

Medical image fusion facilitates the retrieval of complementary information from medical images and has been employed diversely for computer-aided diagnosis of diseases. This paper presents a combination of Principal Component Analysis (PCA) and ridgelet transform as an improved fusion approach for MRI and CT-scan. The proposed fusion approach involves image decomposition using 2D-Ridgelet transform in order to achieve a compact representation of linear singularities. This is followed by application of PCA as a fusion rule to improve upon the spatial resolution. Fusion Factor (FF) and Structural Similarity Index (SSIM) are used as fusion metrics for performance evaluation of the proposed approach. Simulation results demonstrate an improvement in visual quality of the fused image supported by higher values of fusion metrics.

A Non-Linear Approach to ECG Signal Processing using Morphological Filters

Analysis of the Electrocardiogram ECG signals is the pre-requisite for the clinical diagnosis of cardiovascular diseases. ECG signal is degraded by artifacts such as baseline drift and noises which appear during the acquisition phase. The effect of impulse and Gaussian noises is randomly distributed whereas baseline drift generally affects the baseline of the ECG signal; these artifacts induce interference in the diagnosis of cardio diseases. The influence of these artifacts on the ECG signals needs to be removed by suitable ECG signal processing scheme. This paper proposes combination of non linear morphological operators for the noise and baseline drift removal. Non flat structuring elements of varying dimensions are employed with morphological filtering to achieve low distortion as well as good noise removal. Simulation outcomes illustrate noteworthy improvement in baseline drift yielding lower values of MSE and PRD; on the other hand high signal to noise ratios depicts suppression of impulse and Gaussian noises.

Human visual system based unsharp masking for enhancement of mammographic images

Abstract Non-Linear Polynomial Filters (NPF) consists of a schema of linear and quadratic filter components operating as a fusion of low-and high pass filters. NPF has shown distinguished performance when applied for mammogram enhancement. The role has been multifaceted, as there is visual contrast improvement of Region-of-Interest (ROI), i.e. the tumor region as well as those of the surrounding diagnostic features. This paper presents the usage of NPF in design of Non-Linear Unsharp Masking (UM) framework for the enhancement of X-ray mammograms (digital mammographic images). The UM approach presented consists of operational modules namely: edge preserving and contrast enhancement algorithms which are realized using different variants of NPF. Application of Human Visual System (HVS) based adaptive thresholding during contrast enhancement provides for an effective minimization of background noises. The responses of the different modules are then combined using non-linear fusion operators based on an improved logarithmic model of perception and human vision. The obtained enhancement results demonstrate noteworthy improvement in contrast of lesion region together with better visualization of lesion margins and fine details. It has been subjectively as well as objectively shown that the enhancement of the contrast and edges do not introduces unwanted overshoots in the ROI.