More Like this

1. Low Latency Real-Time Seizure Detection Using Transfer Deep Learning

   Khalkhali, Vahid Shawki (December 2021, Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium (SPMB))

   Obeid, Iyad Selesnick (Ed.)

   Scalp electroencephalogram (EEG) signals inherently have a low signal-to-noise ratio due to the way the signal is electrically transduced. Temporal and spatial information must be exploited to achieve accurate detection of seizure events. Most popular approaches to seizure detection using deep learning do not jointly model this information or require multiple passes over the signal, which makes the systems inherently non-causal. In this paper, we exploit both simultaneously by converting the multichannel signal to a grayscale image and using transfer learning to achieve high performance. The proposed system is trained end-to-end with only very simple pre- and post-processing operations which are computationally lightweight and have low latency, making them conducive to clinical applications that require real-time processing. We have achieved a performance of 42.05% sensitivity with 5.78 false alarm per 24 hours on the development dataset of v1.5.2 of the Temple University Hospital Seizure Detection Corpus. On a single core CPU operating at 1.7 GHz, the system runs faster than real-time (0.58 xRT), uses 16 Gbytes of memory, and has a latency of 300 msec. 

   more » « less
2. Fast Automatic Parameter Selection for MRI Reconstruction

   https://doi.org/10.1109/ISBI45749.2020.9098569  

   Toma, Tanjin Taher; Weller, Daniel S. (April 2020, 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI))

   This paper proposes an automatic parameter selection framework for optimizing the performance of parameter-dependent regularized reconstruction algorithms. The proposed approach exploits a convolutional neural network for direct estimation of the regularization parameters from the acquired imaging data. This method can provide very reliable parameter estimates in a computationally efficient way. The effectiveness of the proposed approach is verified on transform-learning-based magnetic resonance image reconstructions of two different publicly available datasets. This experiment qualitatively and quantitatively measures improvement in image reconstruction quality using the proposed parameter selection strategy versus both existing parameter selection solutions and a fully deep-learning reconstruction with limited training data. Based on the experimental results, the proposed method improves average reconstructed image peak signal-to-noise ratio by a dB or more versus all competing methods in both brain and knee datasets, over a range of subsampling factors and input noise levels. 

   more » « less
3. Detecting accelerating eccentric binaries in the LISA band

   https://doi.org/10.1103/PhysRevD.107.043009  

   Xuan, Zeyuan; Naoz, Smadar; Chen, Xian (February 2023, Physical Review D)

   Many gravitational wave (GW) sources in the LISA band are expected to have non-negligible eccentricity. Furthermore, many of them can undergo acceleration because they reside in the presence of a tertiary. Here we develop analytical and numerical methods to quantify how the compact binary's eccentricity enhances the detection of its peculiar acceleration. We show that the general relativistic precession pattern can disentangle the binary's acceleration-induced frequency shift from the chirp-mass-induced frequency shift in GW template fitting, thus relaxing the signal-to-noise ratio requirement for distinguishing the acceleration by a factor of 10 ∼100 . Moreover, by adopting the GW templates of the accelerating eccentric compact binaries, we can enhance the acceleration measurement accuracy by a factor of ∼100 , compared to the zero-eccentricity case, and detect the source's acceleration even if it does not change during the observational time. For example, a stellar-mass binary black hole (BBH) with moderate eccentricity in the LISA band yields an error of the acceleration measurement ∼10-7 m .s−2 for SNR =20 and observational time of 4 yr. In this example, we can measure the BBHs' peculiar acceleration even when it is ∼1 pc away from a 4 ×106M⊙ supermassive black hole. Our results highlight the importance of eccentricity to the LISA-band sources and show the necessity of developing GW templates for accelerating eccentric compact binaries. 

   more » « less
4. Enhancing CMB map reconstruction and power spectrum estimation with convolutional neural networks

   https://doi.org/10.1088/1475-7516/2024/04/041  

   Costanza, Belén; Scóccola, Claudia G; Zaldarriaga, Matías (April 2024, Journal of Cosmology and Astroparticle Physics)

   Abstract The accurate reconstruction of Cosmic Microwave Background (CMB) maps and the measurement of its power spectrum are crucial for studying the early universe. In this paper, we implement a convolutional neural network to apply the Wiener Filter to CMB temperature maps, and use it intensively to compute an optimal quadratic estimation of the power spectrum. Our neural network has a UNet architecture as that implemented in WienerNet, but with novel aspects such as being written inpython 3andTensorFlow 2. It also includes an extra channel for the noise variance map, to account for inhomogeneous noise, and a channel for the mask. The network is very efficient, overcoming the bottleneck that is typically found in standard methods to compute the Wiener Filter, such as those that apply the conjugate gradient. It scales efficiently with the size of the map, making it a useful tool to include in CMB data analysis. The accuracy of the Wiener Filter reconstruction is satisfactory, as compared with the standard method. We heavily use this approach to efficiently estimate the power spectrum, by performing a simulation-based analysis of the optimal quadratic estimator. We further evaluate the quality of the reconstructed maps in terms of the power spectrum and find that we can properly recover the statistical properties of the signal. We find that the proposed architecture can account for inhomogeneous noise efficiently. Furthermore, increasing the complexity of the variance map presents a more significant challenge for the convergence of the network than the noise level does. 

   more » « less
5. FilCorr: Filtered and Lagged Correlation on Streaming Time Series

   https://doi.org/10.1109/icdm50108.2020.00190  

   Zhong, Sheng; Souza, Vinicius M.A.; Mueen, Abdullah (November 2020, 2020 IEEE International Conference on Data Mining (ICDM))

   null (Ed.)

   An essential task on streaming time series data is to compute pairwise correlation across disparate signal sources to identify significant events. In many monitoring applications, such as geospatial monitoring, motion monitoring and critical infrastructure monitoring, correlation is observed at various frequency bands and temporal lags. In this paper, we consider computing filtered and lagged correlation on streaming time series data, which is challenging because the computation must be “in-sync” with the incoming stream for any detected events to be useful. We propose a technique to compute filtered and lagged correlation on streaming data efficiently by merging two individual operations: filtering and cross-correlations. We achieve an order of magnitude speed-up by maintaining frequency transforms over sliding windows. Our method is exact, devoid of sensitive parameters, and easily parallelizable. We demonstrate our technique in a seismic signal monitoring application. 

   more » « less