Suhas Gajre

Novel Method of Using Dynamic Electrical Impedance Signals for Noninvasive Diagnosis of Knee Osteoarthritis

Osteoarthritis (OA) of knee is the most commonly occurring non-fatal irreversible disease, mainly in the elderly population and particularly in female. Various invasive and non-invasive methods are reported for the diagnosis of this articular cartilage pathology. Well known techniques such as X-ray, computed tomography, magnetic resonance imaging, arthroscopy and arthrography are having their disadvantages, and diagnosis of OA in early stages with simple effective noninvasive method is still a biomedical engineering problem. Analyzing knee joint noninvasive signals around knee might give simple solution for diagnosis of knee OA. We used electrical impedance data from knees to compare normal and osteoarthritic subjects during the most common dynamic conditions of the knee, i.e. walking and knee swing. It was found that there is substantial difference in the properties of the walking cycle (WC) and knee swing cycle (KS) signals. In experiments on 90 pathological (combined for KS and WC signals) and 72 normal signals (combined), suitable features were drawn. Then signals were used to classify as normal or pathological. Artificial multilayer feed forward neural network was trained using back propagation algorithm for the classification. On a training data set of 54 signals for KS signals, the classification efficiency for a test set of 54 was 70.37% and 85.19% with and without normalization respectively wrt base impedance. Similarly, the training set of 27 WC signals and test set of 27 signals resulted in 77.78% and 66.67% classification efficiency. The results indicate that dynamic electrical impedance signals have potential to be used as a novel method for noninvasive diagnosis of knee OA

Local gray level S-curve transformation A generalized contrast enhancement technique for medical images

Most medical images suffer from inadequate contrast and brightness, which leads to blurred or weak edges (low contrast) between adjacent tissues resulting in poor segmentation and errors in classification of tissues. Thus, contrast enhancement to improve visual information is extremely important in the development of computational approaches for obtaining quantitative measurements from medical images. In this research, a contrast enhancement algorithm that applies gray-level S-curve transformation technique locally in medical images obtained from various modalities is investigated. The S-curve transformation is an extended gray level transformation technique that results into a curve similar to a sigmoid function through a pixel to pixel transformation. This curve essentially increases the difference between minimum and maximum gray values and the image gradient, locally thereby, strengthening edges between adjacent tissues. The performance of the proposed technique is determined by measuring several parameters namely, edge content (improvement in image gradient), enhancement measure (degree of contrast enhancement), absolute mean brightness error (luminance distortion caused by the enhancement), and feature similarity index measure (preservation of the original image features). Based on medical image datasets comprising 1937 images from various modalities such as ultrasound, mammograms, fluorescent images, fundus, X-ray radiographs and MR images, it is found that the local gray-level S-curve transformation outperforms existing techniques in terms of improved contrast and brightness, resulting in clear and strong edges between adjacent tissues. The proposed technique can be used as a preprocessing tool for effective segmentation and classification of tissue structures in medical images.

ECG Denoising by Modeling Wavelet Sub-Band Coefficients using Kernel Density Estimation

Discrete wavelet transforms are extensively preferred in biomedical signal processing for denoising, feature extraction, and compression. This paper presents a new denoising method based on the modeling of discrete wavelet coefficients of ECG in selected sub-bands with Kernel density estimation. The modeling provides a statistical distribution of information and noise. A Gaussian kernel with bounded support is used for modeling sub-band coefficients and thresholds and is estimated by placing a sliding window on a normalized cumulative density function. We evaluated this approach on offline noisy ECG records from the Cardiovascular Research Centre of the University of Glasgow and on records from the MIT-BIH Arrythmia database. Results show that our proposed technique has a more reliable physical basis and provides improvement in the Signal-to-Noise Ratio (SNR) and Percentage RMS Difference (PRD). The morphological information of ECG signals is found to be unaffected after employing denoising. This is quantified by calculating the mean square error between the feature vectors of original and denoised signal. MSE values are less than 0.05 for most of the cases.

A generalized contrast enhancement approach for knee MR images

Knee Osteoarthritis (OA) is a most prevalent joint disease that can be diagnosed by measuring physiology and morphology of knee joint organs using Magnetic Resonance Imaging (MRI). Measurement of morphological changes in the knee joint organs is a highly challenging task as it requires interpretation and analysis from MR images acquired using different MR pulse sequences. In general, most knee MR images acquired in clinical routines exhibit insignificant tissue contrast and low background luminance in the regions presenting different tissues. This results in the difficulties for further processing and quantitative measurement. In this paper, a method for contrast enhancement in knee MR images acquired using different MR pulse sequences is presented. Local Gray Level Transformation using S-curve technique that has originated from original Gray Level Transform is developed and tested on 6 different knee MR image sequences. The performance of the developed technique is measured by calculating Enhancement Measure (EME), Feature Similarity Index Measure (FSIM) and Absolute Mean Brightness Error (AMBE) and comparing results with existing methods such as Histogram Equalization (THE) and Bi-histogram Based Histogram Equalization (BBHE). Results show significant improvements over existing methods in resolving improper brightness and contrast distribution issues that will contribute to the development of quantitative methods for morphological assessment of knee joint osteoarthritis.

Tropical littoral ambient noise probability density function model based on sea surface temperature

Ambient noise variability in tropical shallow water presents a critical challenge for sonar designers and operators due to site-specific sea surface fluctuations. Sea surface temperature (SST) is a direct measure of energy balance defining the local climate of the region and hence ambient noise characteristics. In this work, an ambient noise probability density function (pdf) model for a spectral band of 3-10 kHz has been designed based on the statistical distribution of SST and validated using real field data. This will enable early ambient noise prediction compared to existing wind speed based models to facilitate structured mitigation strategies for improving sonar performance.

Study of Variation in Ambient Noise with Fluctuations of Surface Parameters for the Indian Ocean Region

Ambient noise variability is a critical challenge encountered by multiple stakeholders, including sonar designers and operators. Among the sources of ambient noise in the ocean, wind related noise has significant impact on sonar performance. The tropical waters in the Indian Ocean Region (IOR), present random fluctuations in the surface parameters, namely the wind speed, surface temperature, wave height, etc. resulting in variations in the ambient noise characteristics. The site-specific surface fluctuations in the tropical regions restrict the possibility of generalized algorithm design to mitigate the ambient noise impact. The work attempts to study the variations in the ambient noise levels corresponding to the fluctuations in the surface parameters. The site-specific behavior of the tropical IOR is demonstrated using surface data available from moored buoy at three distinct locations of the IOR. The analysis methodology can be used to characterize, predict and improve sonar performance, particularly in severe conditions of the tropical IOR.

Validation of Webster ambient noise model for real data in tropical littoral water

Ambient noise has a profound impact on the performance of underwater systems, specifically in the tropical littoral waters. The site specific nature presents its unique challenge and efforts at mitigation have not been effective. Synthetic ambient noise generation has emerged as an effective Modeling and Simulation (M&S) tool for Signal to Noise Ratio (SNR) enhancement. Webster model of random number generation with implicit ambient noise power spectral density structure has been reported to simulate synthetic noise. The power spectral estimation of ambient noise in the tropical littoral waters has always been sub-optimal due to random fluctuations. This work presents the frequency sampling method of FIR filter design for the accurate generation of colored spectra for improved tropical littoral ambient noise generation using the Webster model. This effort focuses on the efficiency of Webster model with expected Power Spectral Density (PSD), kurtosis and the Probability Distribution Function (PDF). The synthetic data generated using the proposed model has been validated using real data recorded by fixed sensors off the west coast of India with a mix of dominance of shipping and wind noise. The recording site has a unique combination of a port being close by, to incorporate dominance of shipping noise in the calm months and the wind noise dominating the spectrum during rough weather condition.

A Local Self-Similarity-Based Vehicle Detection Approach Using Single Query Image

We present a generic vehicle detection approach using a single query image of vehicle to find similar objects in the test image. The proposed method is without any prior training or segmentation of the test image. The approach is based on computing local self-similarity (LSS) descriptors from query and test images that capture local internal geometric layout within the image. Descriptors from query and test image are matched as two-stage process in sliding window framework. In order to exploit usefulness of LSS for generic object detection, we make following contributions: (i) we present few novel ideas to discard the non-informative descriptors to reduce computational expense in feature matching. (ii) We propose a deformation tolerant version of sliding window-based matching framework rather than point-to-point matching. (iii) We also show that selection of landmark points from the query not only makes the algorithm faster but also improves the performance of detection. We evaluate our results on UIUC car dataset, and results clearly outperform earlier training free methods with 91% accuracy.

Analysis of adaptive filtering techniques for fresh water dolphin signals in their natural habitat

Tropical shallow waters typically present poor Signal to Noise Ratio (SNR) for any underwater system. Fresh water habitats experience heavy boat traffics due to significant human encroachments and sharing of habitat with other local species. The Irrawaddy dolphins are known to be facultative species with the significant human presence in these habitats. The boat traffic in their habitat is an important source of noise that degrades their acoustic habitat and even impacts the performance of sonars deployed for monitoring their activities and habitat of these species. The dynamic underwater channel fluctuations of the shallow tropical waters make the design of filters extremely complicated for any SNR enhancement initiative. The dynamic nature of the marine channel originates from the time, frequency and spatial fluctuations of the tropical shallow water environment with varying boundary conditions and multiple boundary interactions as the signal propagates from the source to the receiver. It is unable to track the changes of signal and noise using the fixed coefficient filter in such applications. This work attempts to compare the performance of two adaptive filters to enhance the SNR for dolphin signals in such ambient noise conditions. The two adaptive algorithms LMS (Least Mean Squares) and NLMS (Normalized Least Mean Squares) have been evaluated by comparing the performance parameters such as SNR (Signal-to-Noise Ratio) and MSE (Mean Square Error). The input signal in the work is Irrawaddy dolphin signal in Chilika Lake (19.8450° N, 85.4788° E), that is degraded due high boat traffic of dolphin watching tourist boats. The spectral and the temporal characteristics of recovered signal is verified using the spectrogram method. The simulation study is undertaken using available dolphin click signal and boat noise to be able to identify the precise SNR of the signal at receiver for accurate performance evaluation of the proposed noise mitigation algorithms. This comparative study shows that the NLMS is a better adaptive algorithm for filtering and thus, can improve the performance of a sonar system.

Simulation of colored and non-Gaussian wind noise for tropical shallow waters

Ocean soundscape can be considered to be a sum of identifiable sources and unknown background sound called ambient noise. Ambient noise and its spatial, temporal and spectral variations have a significant impact on performance of sonar systems. These variations are severe in tropical shallow waters and lead to the complexities for prediction and modeling efforts. Underwater acoustic systems suffer sub-optimal performance due to randomness of Signal to Noise Ratio (SNR). Synthetic noise generation has emerged as an effective Modeling and Simulation (M&S) tool for noise estimation and sonar performance enhancement. The Webster model for random number generation along with spectral shaping filters has been widely used to generate ambient noise. In this work we have linked Webster model with the Piggott wind noise model to simulate and validate the ambient noise with real statistics for the tropical shallow water region. The Webster model has been used with specified kurtosis value to generate non-Gaussian sequence and white spectra. Frequency sampling method of FIR filter design is used to generate desired colored spectra. The synthetic data generated using the proposed model has been validated using real data recorded by fixed sensors off the west coast of India based on power spectral density and kurtosis. Results show better spectral coherence of modeled noise with real data for wind speeds close to the threshold, however, there are deviations for higher kurtosis values corresponding to high wind speeds.