Mohammad Ariful Haque

Computer-aided gastrointestinal hemorrhage detection in wireless capsule endoscopy videos

BACKGROUND AND OBJECTIVE Wireless Capsule Endoscopy (WCE) can image the portions of the human gastrointestinal tract that were previously unreachable for conventional endoscopy examinations. A major drawback of this technology is that a large volume of data are to be analyzed in order to detect a disease which can be time-consuming and burdensome for the clinicians. Consequently, there is a dire need of computer-aided disease detection schemes to assist the clinicians. In this paper, we propose a real-time, computationally efficient and effective computerized bleeding detection technique applicable for WCE technology. METHODS The development of our proposed technique is based on the observation that characteristic patterns appear in the frequency spectrum of the WCE frames due to the presence of bleeding region. Discovering these discriminating patterns, we develop a texture-feature-descriptor-based-algorithm that operates on the Normalized Gray Level Co-occurrence Matrix (NGLCM) of the magnitude spectrum of the images. A new local texture descriptor called difference average that operates on NGLCM is also proposed. We also perform statistical validation of the proposed scheme. RESULTS The proposed algorithm was evaluated using a publicly available WCE database. The training set consisted of 600 bleeding and 600 non-bleeding frames. This set was used to train the SVM classifier. On the other hand, 860 bleeding and 860 non-bleeding images were selected from the rest of the extracted images to form the test set. The accuracy, sensitivity and specificity obtained from our method are 99.19%, 99.41% and 98.95% respectively which are significantly higher than state-of-the-art methods. In addition, the low computational cost of our method makes it suitable for real-time implementation. CONCLUSION This work proposes a bleeding detection algorithm that employs textural features from the magnitude spectrum of the WCE images. Experimental outcomes backed by statistical validations prove that the proposed algorithm is superior to the existing ones in terms of accuracy, sensitivity, specificity and computational cost.

Epilepsy and seizure detection using statistical features in the Complete Ensemble Empirical Mode Decomposition domain

In this paper, we introduce Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) to devise an effective feature extraction scheme for physiological signal analysis. Unlike its predecessors- Empirical Mode Decomposition and Ensemble Empirical Mode Decomposition, CEEMDAN resolves mode mixing problem and gives better spectral separation of the modes. To demonstrate the effectiveness of CEEMDAN based features, we apply CEEMDAN to propose an automatic epileptic seizure detection algorithm. In this work, various statistical features are extracted from the EEG signal segments decomposed by CEEMDAN and seizure classification is performed using artificial neural network. The efficacy of our feature extraction scheme is validated by statistical and graphical analyses. The overall performance of our seizure detection scheme as compared to the state-of-the-art ones is also promising.

Robust Speech Dereverberation Based on Blind Adaptive Estimation of Acoustic Channels

This paper addresses the problem of speech dereverberation considering a noisy and slowly time-varying environment. The proposed multimicrophone speech dereverberation model utilizes the estimated acoustic impulse responses (AIRs) to dereverberate the speech as well as improve the signal-to-noise ratio without a priori information about the AIRs, location of the source and microphones, or statistical properties of the speech/noise, which are some common assumptions in the related literature. The received noisy signals are filtered through an eigenfilter which improves the power of the speech signal as compared to that of the additive noise. The eigenfilter is efficiently computed avoiding the tedious Cholesky decomposition, solely from the estimates of the AIRs. The design of the eigenfilter also incorporates a frequency domain constraint that improves the quality of the speech signal, resists spectral nulls in addition to improving the signal-to-noise ratio (SNR). A zero-forcing equalizer (ZFE) is used to dereverberate the speech signal by eliminating the distortion caused by the AIRs as well as the eigenfilter. The ZFE is implemented in block-adaptive form which makes the proposed technique suitable for speech dereverberation in a time-varying condition. The simulation results verify the superior performance of the proposed method as compared to the state-of-the-art dereverberation techniques in terms of log-likelihood ratio (LLR), segSNR, weighted spectral slope (WSS), and perceptual evaluation of speech quality (PESQ).

Real-Time Robust Heart Rate Estimation From Wrist-Type PPG Signals Using Multiple Reference Adaptive Noise Cancellation

Heart rate (HR) monitoring using photoplethysmographic (PPG) signals recorded from wearers’ wrist greatly facilitates design of wearable devices and maximizes user experience. However, placing PPG sensors in wrist causes much stronger and complicated motion artifacts (MA) due to loose interface between sensors and skin. Therefore, developing robust HR estimation algorithms for wrist-type PPG signals has significant commercial values. In this paper, we propose a robust HR estimation algorithm for wrist-type PPG signals using multiple reference adaptive noise cancellation (ANC) technique—termed here as “MURAD.” The main challenge of using ANC for MA reduction is to devise a qualified reference noise signal (RNS) to the adaptive filter. We propose a novel solution by using four RNSs, namely, the three-axis accelerometer data and the difference signal between the two PPG signals. For each RNS, we get a different version of the cleaned PPG signal. Then, a set of probable HR values is estimated using all of the cleaned PPG signals, and then, the value that is closest to the estimated HR of the previous time window is chosen to be the HR estimate of the current window. Then, some peak verification techniques are employed to ensure accurate HR estimations. The proposed technique gives lower average absolute error compared to state-of-the art methods. So, MURAD method provides a promising solution to the challenge of HR monitoring using PPG in wearable devices during severe MA conditions.

Non-intrusive objective evaluation of speech quality in noisy condition

It is very difficult, if not impossible, to obtain a clean reference signal of a noisy speech recorded in a practical environment. As a result, intrusive methods that evaluate the quality of speech signal with the help of a clean reference signal has little value in real world applications. In this paper, we investigate the effectiveness of data-driven non-intrusive method for assessing quality of speech without using clean reference signal. In the proposed method, a support vector machine based classifier is trained using a labelled dataset and then the classifier provides speech assessment rating on unknown speech signals. The obtained results have been evaluated against the intrusive PESQ score. The results indicate that the proposed technique performs better than the state-of-the-art non-intrusive methods on the same test data set.

Prevalence of Health Diseases among Bangladeshi Tannery Workers and associated Risk factors with Workplace Investigation

Background: The study aimed to estimate the prevalence of health hazards through analyzing different diseases among tannery workers and to identify risks factors of these diseases in tannery workers in Dhaka, Bangladesh. Methods: A cross sectional study was conducted from May 2016 to July 2016 on tannery workers of the industrial area Hazaribagh, Dhaka. Data collection was done among 276 tannery workers engaged in different tanneries located at the area. Face to face interview was performed using a questionnaire and skin diseases faced by the workers were identified by trained medical students. Results: The prevalence of diseases was found as gastrointestinal problem 71.7%, Diarrhea 71.7%, Blood pressure 52.2%, Asthma 49.9%, Eye problem 46.7%. Skin disease prevailed in the following order: Scabies 73.9%, Nail discoloration 69.6%, Urticaria 59.7%, Miliria and foliculities 56.5%. Again negligence of using personal protective equipment named safety boots, gloves and googles have significant influence (p<0.001) on higher prevalence of asthma, eye problems and nail discoloration. Moreover, working areas of pre-tanning and tanning have significant (P<0.001) association with the higher prevalence of scabies and miliria and foliculities. Conclusion: Prevalence of diseases among the tannery workers are very high and is extremely associated with different working areas of leather processing and the lack of proper PPE (Personal protective equipment) using. The risk factors responsible for these health hazards can be eliminated through improving the overall working condition and ensuring the necessary protective regulatory for the Bangladeshi tannery workers.

Multi-step level set method for segmentation of overlapping cervical cells

Automatic segmentation of the overlapping cervical cells is one of the most challenging problems in the medical image analysis. This paper presents a novel multi-step level set (LVS) method for segmenting cytoplasm and nuclei from overlapping cells in a single EDF image produced from Pap smear images of multi-layer cervical cell volumes. The first step segments the clump consisting of free or overlapping cells using a region-and edge-based level set method on the Gaussian filtered image. The second step segments the nuclei by a multi-step level set method from the original image. And finally, the most critical step of cytoplasm segmentation is done using level set method optimized by criteria such as curvature of the cytoplasm, duration of retaining the segmented area of the cytoplasm, edge information and a speed regulator depends on the homogeneity of the cell. The performance of the proposed algorithm is evaluated on the real cervical cell image provided by the second overlapping cervical cytology image segmentation challenge at ISBI 2015. Clump detection shows good result in spite of the weak clump boundary. Nuclei detection also shows good result in spite of congestion of nuclei. We are also able to segment two overlapping cells in the real Pap smear image.

Resolution enhancement of CT images using multiple low resolution sub-pixel shifted images

Sub-pixel shifted low resolution images contain new information about the same scene which can be exploited to generate high resolution (HR) image. In computed tomography (CT), HR image reconstruction using limited projection data is always desired but it requires iterative techniques that involve high computational cost. In this paper, we propose a computationally efficient and noise robust HR image reconstruction method using multiple low resolution images. The work is divided into two stages: first we demonstrate how to reconstruct a number of low resolution images utilizing the same projection data and then we employ standard interpolation/super resolution technique to obtain an HR image. The method is not only computationally efficient but also improves signal to noise ratio of the reconstructed images. The simulation results show that the proposed method can produce better images compared to the state-of-the-art techniques.

Utterance-based speech dereverberation using blind channel estimation and multichannel equalization

Speech dereverberation considering noisy environment as well as speakers movement is a challenging task. In this paper, we present an utterance-based noise-robust speech dereverberation technique that is suitable for non-stationary speaker. The acoustic impulse responses (AIRs) between the speaker and microphone array are estimated using the spectrally constrained frequency-domain least-mean-squares (LMS) algorithm. The AIRs are then equalized using the iterative multiple-input/output inverse theorem (MINT). It is assumed that the speaker stays still within an utterance, however, the speaker changes his/her position between the utterances. The simulation experiments conducted in various reverberant environment and speakers position demonstrate that the proposed method can satisfactorily improve the perceptual quality of the noisy reverberated speech.

Fast adaptive frequency-domain mutually referenced equalizers for SIMO channels with noise

This paper deals with improving the performance of reduced mutually referenced equalizers for direct blind equalization of SIMO FIR channels in the presence of additive noise. Here we propose a noise-robust optimal-step-size frequency-domain LMS algorithm for estimating the equalizer coefficients. The proposed technique resolves the step-size ambiguity of the LMS algorithm, gives faster convergence speed as compared to the time-domain counterpart. Computer simulation results are presented to show its improved performance for blind adaptive equalization.