Deep learning (DL) algorithms are prevalent in radiology as workflow assistants and as modality enhancements. Magnetic resonance imaging (MRI), computerized tomography (CT), diagnostic imaging (DI), ultrasound (US), and positron emission tomography (PET) are modalities that benefit from the DL algorithms and shorter exam times or greater image accuracy. Faster scan time is achieved by the signal to noise ratio (SNR). The distinction is that the technology can enhance images beyond the original resolution from the modality or shorten exam time and rebuild the image quality through SNR algorithms back to approximately the original standard of care (SOC) image. While artificial intelligence signal to noise ratio algorithms (AI-SNR) can enhance an image to a greater accuracy, shorter exam times are a measurable component of a return on investment (ROI) in calculating modality utilization. The algorithm-derived images may have visual variations that are not found on normally acquired original images. The research focused on DL-SNR images on a three-tesla magnetic resonance imaging (3.0T MRI) unit, a high resolution MRI deployment in the industry. The primary research question for this research study is: What are radiologists' perceptions in regard to image quality and increased utilization due to vendor provided DL-SNR on 3.0T MRI? This will be an exploratory qualitative research study using detailed interviews with fellowship-trained radiologists that are using AI-SNR in 3.0T MRI and shortened exam time protocols. Fifteen interviews were conducted. The interview transcriptions were coded using ATLAS.ti to identify common themes and sub-themes in the radiologists' perceptions of DL-SNR imaging. This paper assumes the reader has an adequate understanding of deep learning and radiology processes. The interviews included discussions on key elements on image quality, workflow, reimbursement, legal concerns, and radiologist workload. Issues were identified and potential solut (open full item for complete abstract)

This work describes a fully analog ECG motion artifacts (MAs) and baseline wander elimination circuit that can be incorporated in the analog sensor frontend. In the proposed method, the R-peaks, as useful components of ECG signals, are detected by a high pass filter (HPF) and excluded from the moving average input. By linearly interpolating the down-sampled moving average output, the baseline wander can be effectively detected. The final output is generated by subtracting the extracted baseline wander from the corrupted ECG waveforms. Owing to various types of switch capacitor integrated circuits including the Biquad filter, integrator and double sampling sample and hold (S/H) circuits for realizing DC offset and abrupt changes removal, HPF and moving average, linear interpolation and delay chain circuits, this method can be implemented fully on-chip. Therefore, the power consumption and chip area are drastically reduced compared to existing schemes, thus, can be suitable for using in long-term ECG monitoring devices. The proposed algorithm is implemented on the single chip (utilizing 0.18-μm CMOS technology with 1.8-V power supply) and verified by on-body testing over 24 subjects within the age group of 10 to 55 for different types of motion artifacts due to various activities including walking, texting, sleeping, and intentionally touching the skin electrodes. The measurement results show signal-to-noise ratio (SNR) improvement of almost 16-dB and on average 10% improvement in delta percentage root-mean-square difference (ΔPRD), where the power consumption of the chip is only 6.6-μW with core area of 0.85×2.16 mm2.

Speech is essential for human communication as it not only delivers messages but also expresses emotions. In reality, speech is often corrupted by background noise and room reverberation. Perceiving speech in low signal-to-noise ratio (SNR) conditions is challenging, especially for hearing-impaired listeners. Therefore, we are motivated to develop speech separation algorithms to improve intelligibility of noisy speech. Given its many applications, such as hearing aids and robust automatic speech recognition (ASR), speech separation has been an important problem in speech processing for decades. Speech separation can be achieved by estimating the ideal binary mask (IBM) or ideal ratio mask (IRM). In a time-frequency (T-F) representation of noisy speech, the IBM preserves speech-dominant T-F units and discards noise-dominant ones. Similarly, the IRM adjusts the gain of each T-F unit to suppress noise. As such, speech separation can be treated as a supervised learning problem where one estimates the ideal mask from noisy speech. Three key components of supervised speech separation are learning machines, acoustic features and training targets. This supervised framework has enabled the treatment of speech separation with powerful learning machines such as deep neural networks (DNNs). For any supervised learning problem, generalization to unseen conditions is critical. This dissertation addresses generalization of supervised speech separation. We first explore acoustic features for supervised speech separation in low SNR conditions. An extensive list of acoustic features is evaluated for IBM estimation. The list includes ASR features, speaker recognition features and speech separation features. In addition, we propose the Multi-Resolution Cochleagram (MRCG) feature to incorporate both local information and broader spectrotemporal contexts. We find that gammatone-domain features, especially the proposed MRCG features, perform well for supervised speech separation at (open full item for complete abstract)

Automatic modulation classifiers (AMC) are one of the basic building blocks of electronic warfare receivers and cognitive radios. Although many research papers on AMC algorithms have been published, very few results on their implementation are available. This thesis presents a feature-based AMC built upon a software-defined radio platform. The developed AMC can detect signals over a broad spectrum and classify the modulation used. The modulation schemes considered in this thesis are amplitude modulation (AM), frequency modulation (FM), phase-shift keying (PSK), and quadrature amplitude modulation (QAM). Experimental results demonstrate the validity of the developed AMC algorithm and its implementation.

Cooperative relaying is a promising technique to enhance the reliability and data-rate of wireless networks. Among different cooperative relaying schemes, the half-duplex non-orthogonal amplify-and-forward (NAF) protocol is popular due to its low implementation complexity and performance advantages. This thesis investigates precoder design for a cooperative half-duplex single-relay NAF system from an information theoretic point of view using a novel mutual information-based design criterion. The first part of the thesis considers the design of a 2x2 precoder for the NAF half duplex single relay network in the presence of the direct link using mutual information (MI) as the main performance metric. Different from precoder design methods using pairwise error probability (PEP) analysis, which are valid only at high signal-to-noise ratios (SNR), the proposed precoder design can apply to any SNR region, which is of more interest from both information-theoretic and practical points of view. A MI-based criterion is developed for a cooperative frame length of 2, which corresponds to the case of using a 2x2 precoder. The design criterion is established in a closed-form, which can be helpful in finding an optimal precoder. Then it is analytically shown that a good precoder should have all entries that are equal in magnitude, which is different from the optimal precoders obtained thus far using the conventional PEP criterion. Simulation results indicate that the proposed class of precoders outperform the existing precoders in terms of the mutual information performance. The second part of the thesis extends the precoder design to an arbitrary block length of 2T. In general, for this case, a precoder of size 2Tx2T needs to be considered to optimize the MI performance. Similar to the 2 × 2 precoder design, a MI-based design criterion is first established. While the criterion can be expressed in closed form, the design of optimal precoders in this case is not possible, (open full item for complete abstract)

We introduce an improved 2D adaptive smoothing algorithm for noise removal and feature preservation. Comparing to the original 2D adaptive smoothing algorithm, this new algorithm is also based on the novel idea of utilizing contextual discontinuity and local discontinuity jointly to detect and distinguish edges and noise. The new algorithm improves the main concept – contextual discontinuity by introducing a novel homogeneity region definition with a corresponding method for contextual discontinuity measurement. Comparing to the original algorithm and other smoothing algorithms, the improved algorithm can preserve edges more effectively while removing noise.The improved 2D algorithm has been implemented and extensive experiments have been carried out to compare the algorithm to the original algorithm and other smoothing strategies to quantitatively demonstrate improvement in performance. Measurements are applied to evaluate the noise removal and edge preservation performance. Simulation results show that this improved algorithm has a superior performance over both the original algorithm and other popular smoothing strategies in noise removal as well as feature preservation.

In a natural environment, speech often occurs simultaneously with acoustic interference. Many applications, such as automatic speech recognition and telecommunication, require an effective system that segregates speech from interference in the monaural (one-microphone) situation. While this task of monaural speech segregation has proven to be very challenging, human listeners show a remarkable ability to segregate an acoustic mixture and attend to a target sound, even with one ear. This perceptual process is called auditory scene analysis (ASA). Research in ASA has inspired considerable effort in constructing computational ASA (CASA) based on ASA principles. Current CASA systems, however, face a number of challenges in monaural speech segregation. This dissertation presents a systematic and extensive effort in developing a CASA system for monaural speech segregation that addresses several major challenges. The proposed system consists of four stages: Peripheral analysis, feature extraction, segmentation, and grouping. In the first stage, the system decomposes the auditory scene into a time-frequency representation via bandpass filtering and time windowing. The second stage extracts auditory features corresponding to ASA cues, such as periodicity, amplitude modulation, onset and offset. In the third stage, the system segments an auditory scene based on a multiscale analysis of onset and offset. The last stage includes an iterative algorithm that simultaneously estimates the pitch of a target utterance and segregates the voiced target based on a pitch estimate. Finally, our system sequentially groups voiced and unvoiced portions of the target speech for non-speech interference, and this grouping task is performed using feature-based classification. Systematic evaluation shows that the proposed system extracts a majority of target speech without including much interference. Extensive comparisons demonstrate that the system has substantially advanced the state-of-the-ar (open full item for complete abstract)

The cross-range resolution of a laser radar (ladar) system can be improved by synthesizing a large aperture from multiple smaller sub-apertures. This aperture synthesis requires a coherent combination of the sub-apertures; that is, the sub-apertures must be properly phased and placed with respect to each other. One method that has been demonstrated in the literature to coherently combine the sub-apertures is to cross-correlate the speckle patterns imaged in overlapping regions. This work investigates the effect of low signal to noise ratio (SNR) on an efficient speckle cross-correlation registration algorithm with sub-pixel accuracy. Specifically, the algorithms ability to estimate relative piston and tilt errors between sub-apertures at low signal levels is modeled and measured. The effects of these errors on image quality are examined using the modulation transfer function (MTF) as a metric. The results demonstrate that in the shot noise limit, with signal levels as low as about 0.02 signal photoelectrons per pixel in a typical CCD, the registration algorithm estimates relative piston and tilt accurately to within 0.1 radians of true piston and 0.1 waves of true tilt. If the sub-apertures are not accurately aligned in the synthetic aperture, then the image quality degrades as the number of sub-apertures increases. The effect on the MTF is similar to the effects due to defocus aberrations.

MRI pulse sequences are used to acquire spatial-frequency data (k-space) by using linear magnetic field gradients to encode a k-space trajectory. Common k-space trajectories including rectilinear typically limit sampling to periods of constant gradient amplitude resulting in uniformly-sampled data during the acquisition period. However, these trajectories often do not acquire data during gradient ramp times or during read gradient dephase/rephase lobes. In this work, data acquisition and image reconstruction techniques were developed to improve the effective utilization of these times to improve image quality. Continuous sampling (CS) is defined as data collected anytime after RF excitation. A Cartesian CS acquisition was developed by extending the acquisition window over the entire balanced and isolated read gradient. The gradients were designed to sample each k-space pixel at least twice to essentially accomplish two averages in a single acquisition. Cartesian CS resulted in over 40% improvement in SNR for a true-FISP sequence relative to the traditional Cartesian acquisition in phantom and asymptomatic volunteer imaging studies. Non-Cartesian CS sequences including the BOWTIE trajectory and unequal gradient amplitude CS were developed for faster CS sequences with similar SNR gains. For the unequal gradient amplitude case a range of SNR gain of 19 to 50% was seen. The BOWTIE trajectory was accomplished by overlapping PE and read gradients and resulted in a 35% increase in SNR compared to the traditional Cartesian. Both resulted in a reduction of imaging time at the cost of SNR compared to the original CS sequence. A Cartesian CS Dixon fat/water separation techniques was developed by applying the Dixon echo time variation between CS echoes. The redundant data collected during CS produced two Dixon images in one acquisition. The CS Dixon resulted in equivalent fat suppression in water images in approximately half the time compared to a 2PD acquisition. A novel 2D tr (open full item for complete abstract)

Carpal Tunnel Syndrome (CTS) is a musculoskeletal disorder characterized by the compression of an enlarged or inflamed median nerve as it passes through the carpal tunnel and deep to the flexor retinaculum. CTS is one of the most common entrapment syndromes of the upper limbs, with hundreds of thousands of new cases of CTS reported by the Centers for Disease Control and Prevention in the United States every year (CDC,2012). Presently, according to the ACR Appropriateness Criteria, Magnetic Resonance Imaging (MRI) has a rating of nine out of ten as the best choice for diagnosing CTS in patients with persistent wrist pain after the initial radiograph. The Appropriateness Criteria also lists musculoskeletal (MSK) sonography with a rating of one out of ten; however, qualities such as accessibility, cost effectiveness, being less invasive, relatively painless, time effectiveness, as well as providing real time imaging, may provide additional information in conjunction with Magnetic Resonance. The purpose of this study was to evaluate the significance, if any, through quantitative analysis of the median nerve, between Magnetic Resonance and Sonography in the acute stages of CTS. Imaging was performed on Maccaca fascularis monkeys at baseline, working, and recovery intervals. The data was collected from two independent, blinded researchers, one certified in Magnetic Resonance, the other certified in Sonography. Although each study demonstrated no conclusive comparison between MRI and Sonography in the evaluation of the median nerve, the information gained regarding study protocol is invaluable to provide feedback to design a higher level clinical study. MSK sonography may be a useful tool in combination with MRI, to diagnose CTS, with minimal discomfort to the patient. More research needs to be conducted in the acute stages of CTS before the patient reaches the advanced, symptomatic stages, in the form of a clinical human study.

A hybrid filtering method called Karhunen Loeve Transform-Wavelet (KW) filtering is presented to de-noise dynamic cardiac magnetic resonance images that simultaneously takes advantage of the intrinsic spatial and temporal redundancies of real-time cardiac cine. This new image filtering technique combines two well-established methods: temporal Karhunen-Loeve transform (KLT) and spatial adaptive wavelet filtering. KW filtering has four steps: 1. Apply KLT along the temporal direction, generating a series of “eigenimages”. Because of the high temporal correlations, most of the energy is concentrated into a few eigenimages. 2. Marcenko-Pastur (MP) law is used to identify and discard the noise-only eigenimages; 3. 2-D spatial wavelet filter with adaptive threshold is applied to each eigenimage. An adaptive threshold is used to define the wavelet filter strength for each of the eigenimages based on the noise variance and standard deviation of the signal, resulting in stronger filtering of the eigenimages that primarily contain noise. 4. Apply the inverse KLT to the filtered eigenimages to generate a new series of cine images with reduced image noise. KW filter was compared with 2 other filters – Spatial Wavelet filter and Temporal KLT filter in terms of SNR gain and edge sharpness. For four volunteer data acquired using rate 5 acceleration, KW filter showed an SNR gain of 98%. For a matched value of SNR gain between KW filter, Wavelet filter and KLT filter, KW filter preserved 93.83% of original image sharpness while Wavelet filter and KLT filter preserved 82.23% and 88.05% respectively.