Focused analysis of RNFL decay in glaucomatous eyes using circular statistics on high-resolution OCT data.

We generated Optical Coherence Tomography (OCT) data of much higher resolution than usual on retinal nerve fiber layer (RNFL) thickness of a given eye. These consist of measurements made at hundreds of angular-points defined on a circular coordinate system. Traditional analysis of OCT RNFL data does not utilize insightful characteristics such as its circularity and granularity for common downstream applications. To address this, we present a new circular statistical framework that defines an Angular Decay function and thereby provides a directionally precise representation of an eye with attention to patterns of focused RNFL loss. By applying to a clinical cohort of Asian Indian eyes, the generated circular data were modeled with a finite mixture of von Mises distributions, which led to an unsupervised identification in different age-groups of recurrent clusters of glaucomatous eyes with distinct directional signatures of RNFL decay. New indices of global and local RNFL loss were computed for comparing the structural differences between these glaucoma clusters across the age-groups and improving classification. Further, we built a catalog of directionally precise statistical distributions of RNFL thickness for the said population of normal eyes as stratified by their age and optic disc size.

Lightweight deep convolutional neural network for background sound classification in speech signals.

Recognizing background information in human speech signals is a task that is extremely useful in a wide range of practical applications, and many articles on background sound classification have been published. It has not, however, been addressed with background embedded in real-world human speech signals. Thus, this work proposes a lightweight deep convolutional neural network (CNN) in conjunction with spectrograms for an efficient background sound classification with practical human speech signals. The proposed model classifies 11 different background sounds such as airplane, airport, babble, car, drone, exhibition, helicopter, restaurant, station, street, and train sounds embedded in human speech signals. The proposed deep CNN model consists of four convolution layers, four max-pooling layers, and one fully connected layer. The model is tested on human speech signals with varying signal-to-noise ratios (SNRs). Based on the results, the proposed deep CNN model utilizing spectrograms achieves an overall background sound classification accuracy of 95.2% using the human speech signals with a wide range of SNRs. It is also observed that the proposed model outperforms the benchmark models in terms of both accuracy and inference time when evaluated on edge computing devices.

Circular functional analysis of OCT data for precise identification of structural phenotypes in the eye.

Progressive optic neuropathies such as glaucoma are major causes of blindness globally. Multiple sources of subjectivity and analytical challenges are often encountered by clinicians in the process of early diagnosis and clinical management of these diseases. In glaucoma, the structural damage is often characterized by neuroretinal rim (NRR) thinning of the optic nerve head, and other clinical parameters. Baseline structural heterogeneity in the eyes can play a key role in the progression of optic neuropathies, and present challenges to clinical decision-making. We generated a dataset of Optical Coherence Tomography (OCT) based high-resolution circular measurements on NRR phenotypes, along with other clinical covariates, of 3973 healthy eyes as part of an established clinical cohort of Asian Indian participants. We introduced CIFU, a new computational pipeline for CIrcular FUnctional data modeling and analysis. We demonstrated CIFU by unsupervised circular functional clustering of the OCT NRR data, followed by meta-clustering to characterize the clusters using clinical covariates, and presented a circular visualization of the results. Upon stratification by age, we identified a healthy NRR phenotype cluster in the age group 40-49 years with predictive potential for glaucoma. Our dataset also addresses the disparity of representation of this particular population in normative OCT databases.

A low-complexity and robust minimum variance beamformer for ultrasound imaging systems using beamspace dominant mode rejection.

Minimum variance beamformer (MVB) has a high computational complexity that is mainly due to the inversion of an L×L covariance matrix involved during weight vector estimation, where L is the length of the subarray. In this work an attempt is made to reduce the computational complexity as well as increase the robustness against signal mismatch. The computational complexity is reduced by projecting the element-space data on to beamspace domain and then using dominant mode rejection on the beamspace covariance matrix (BCM). This reduces the dimension of covariance matrix and also eliminates the matrix inversion thereby reducing the computational complexity. Further, a closeness factor is introduced to determine the interference components that have to be suppressed, leading to increased robustness against signal mismatch. Performance of the proposed method has been evaluated on both simulated and experimental datasets. Results indicate that the proposed beamformer has a lateral resolution of 0.07 mm and a contrast resolution of 0.80, which are comparable to that of MVB which has a lateral and contrast resolution of 0.10 mm and 0.78, that too with a 12-fold reduction in computational complexity. The robustness of the peak magnitude estimate and spatial resolution of the beamformer with respect to error in estimation of sound velocity in the medium have also been evaluated. The variation in lateral resolution of proposed beamformer and MVB is approximately 1.21 mm and 1 mm. Further, the proposed beamformer has a maximum deviation in peak magnitude estimate of 0.7 dB whereas that of MVB is 2.5 dB, thus indicating the increased robustness of proposed method in peak magnitude estimate. Overall, the proposed beamformer has a 12-fold lower computational complexity compared to MVB with additional flexibility to increase the robustness against error in sound velocity.

A low power, low noise Programmable Analog Front End (PAFE) for biopotential measurements.

A low power Programmable Analog Front End (PAFE) for biopotential measurements is presented in this paper. The PAFE circuit processes electrocardiogram (ECG), electromyography (EMG) and electroencephalogram (EEG) signals with higher accuracy. It consists mainly of improved transconductance programmable gain instrumentational amplifier (PGIA), programmable high pass filter (PHPF), and second order low pass filter (SLPF). A 15-bit programmable 5-stage successive approximation analog-to-digital converter (SAR-ADC) is implemented for improving the performance, whose power consumption is reduced due to multiple stages and by OTA/Comparator sharing technique between the stages. The power consumption is further reduced by operating the analog portion of PAFE on 0.5V supply voltage and digital portion on 0.3V supply voltage generated internally through a voltage regulator. The proposed low power PAFE has been fabricated in 180nm standard CMOS process. The performance parameters of PAFE in 15-bit mode are found to be, gain of 31-70 dB, input referred noise of 1.15 µVrms, CMRR of 110 dB, PSRR of 104 dB, and signal-to-noise distortion ratio (SNDR) of 83.5dB. The power consumption of the design is 1.1 µW @ 0.5 V supply voltage and it occupies a core silicon area of 1.2 mm2.

A smart phone/tablet based mobile health care system for developing countries.

Increasing cost of health care in developing countries is placing heavy financial burden on its populations. With the advent of mobile and tablet technologies however, it is possible to reduce this burden to some extent through tele-healthcare. In this paper, authors describe their effort to design portable diagnostic devices that can communicate to smart phones and tablets there by making tele-healthcare possible. A possible architecture of their model is presented and components thereof discussed.

Uniform approximation of Gaussian wavelet for biomedical signal processing in analog domain.

Signal processing in analog domain is favorable when power consumption is a critical design constraint. Continuous Wavelet Transform (CWT), which is increasingly being used in characterizing biomedical signals, when implemented in analog domain consumes less power provided the mother wavelet is properly approximated. This paper presents an approximation of Gaussian wavelet by making use of the Uniform approximation. Simulations of the approximated wavelet and the actual wavelet in MATLAB are performed and the results discussed. Simulations show that (i) approximation obtained closely matches the mother wavelet chosen and (ii) a stable approximation which helps in physical realization using any circuit design methodology.