We present the first machine learning-based autonomous hyperspectral neutron computed tomography experiment performed at the Spallation Neutron Source. Hyperspectral neutron computed tomography allows the characterization of samples by enabling the reconstruction of crystallographic information and elemental/isotopic composition of objects relevant to materials science. High quality reconstructions using traditional algorithms such as the filtered back projection require a high signal-to-noise ratio across a wide wavelength range combined with a large number of projections. This results in scan times of several days to acquire hundreds of hyperspectral projections, during which end users have minimal feedback. To address these challenges, a golden ratio scanning protocol combined with model-based image reconstruction algorithms have been proposed. This novel approach enables high quality real-time reconstructions from streaming experimental data, thus providing feedback to users, while requiring fewer yet a fixed number of projections compared to the filtered back projection method. In this paper, we propose a novel machine learning criterion that can terminate a streaming neutron tomography scan once sufficient information is obtained based on the current set of measurements. Our decision criterion uses a quality score which combines a reference-free image quality metric computed using a pre-trained deep neural network with a metric that measures differences between consecutive reconstructions. The results show that our method can reduce the measurement time by approximately a factor of five compared to a baseline method based on filtered back projection for the samples we studied while automatically terminating the scans.

Hyperspectral neutron tomography is an effective method for analyzing crystalline material samples with complex compositions in a non-destructive manner. Since the counts in the hyperspectral neutron radiographs directly depend on the neutron cross-sections, materials may exhibit contrasting neutron responses across wavelengths. Therefore, it is possible to extract the unique signatures associated with each material and use them to separate the crystalline phases simultaneously.We introduce an autonomous material decomposition (AMD) algorithm to automatically characterize and localize polycrystalline structures using Bragg edges with contrasting neutron responses from hyperspectral data. The algorithm estimates the linear attenuation coefficient spectra from the measured radiographs and then uses these spectra to perform polycrystalline material decomposition and reconstructs 3D material volumes to localize materials in the spatial domain. Our results demonstrate that the method can accurately estimate both the linear attenuation coefficient spectra and associated reconstructions on both simulated and experimental neutron data.

Neutron computed tomography (nCT) is a 3D characterization technique used to image the internal morphology or chemical composition of samples in biology and materials sciences. A typical workflow involves placing the sample in the path of a neutron beam, acquiring projection data at a predefined set of orientations, and processing the resulting data using an analytic reconstruction algorithm. Typical nCT scans require hours to days to complete and are then processed using conventional filtered back-projection (FBP), which performs poorly with sparse views or noisy data. Hence, the main methods in order to reduce overall acquisition time are the use of an improved sampling strategy combined with the use of advanced reconstruction methods such as model-based iterative reconstruction (MBIR). In this paper, we propose an adaptive orientation selection method in which an MBIR reconstruction on previously-acquired measurements is used to define an objective function on orientations that balances a data-fitting term promoting edge alignment and a regularization term promoting orientation diversity. Using simulated and experimental data, we demonstrate that our method produces high-quality reconstructions using significantly fewer total measurements than the conventional approach.

The study of human reaction time (RT) is invaluable not only to understand the sensory-motor functions but also to translate brain signals into machine comprehensible commands that can facilitate augmentative and alternative communication using brain-computer interfaces (BCI). Recent developments in sensor technologies, hardware computational capabilities, and neural network models have significantly helped advance biomedical signal processing research. This study is an attempt to utilize state-of-the-art resources to explore the relationship between human behavioral responses during perceptual decision-making and corresponding brain signals in the form of electroencephalograms (EEG). In this paper, a generalized 3D convolutional neural network (CNN) architecture is introduced to estimate RT for a simple visual task using single-trial multi-channel EEG. Earlier comparable studies have also employed a number of machine learning and deep learning-based models, but none of them considered inter-channel relationships while estimating RT. On the contrary, the use of 3D convolutional layers enabled us to consider the spatial relationship among adjacent channels while simultaneously utilizing spectral information from individual channels. Our model can predict RT with a root mean square error of 91.5 ms and a correlation coefficient of 0.83. These results surpass all the previous results attained from different studies.Clinical relevance Novel approaches to decode brain signals can facilitate research on brain-computer interfaces (BCIs), psychology, and neuroscience, enabling people to utilize assistive devices by root-causing psychological or neuromuscular disorders.

The estimation of the visual stimulus-based reaction time (RT) using subtle and complex information from the brain signals is still a challenge, as the behavioral response during perceptual decision making varies inordinately across trials. Several investigations have tried to formulate the estimation based on electroencephalogram (EEG) signals. However, these studies are subject-specific and limited to regression-based analysis. In this paper, for the first time to our knowledge, a generalized model is introduced to estimate RT using single-trial EEG features for a simple visual reaction task, considering both regression and classification-based approaches. With the regression-based approach, we could predict RT with a root mean square error of 111.2 ms and a correlation coefficient of 0.74. A binary and a 3-class classifier model were trained, based on the magnitude of RT, for the classification approach. Accuracy of 79% and 72% were achieved for the binary and the 3-class classification, respectively. Limiting our study to only high and low RT groups, the model classified the two groups with an accuracy of 95%. Relevant EEG channels were evaluated to localize the part of the brain significantly responsible for RT estimation, followed by the isolation of important features.Clinical relevance- Electroencephalogram (EEG) signals can be used in Brain-computer interfaces (BCIs), enabling people with neuromuscular disorders like brainstem stroke, amyotrophic lateral sclerosis, and spinal cord injury to communicate with assistive devices. However, advancements regarding EEG signal analysis and interpretation are far from adequate, and this study is a step forward.

Multiplexed deep neural networks (DNN) have engendered high-performance predictive models gaining popularity for decoding brain waves, extensively collected in the form of electroencephalogram (EEG) signals. In this paper, to the best of our knowledge, we introduce a first-ever DNN-based generalized approach to estimate reaction time (RT) using the periodogram representation of single-trial EEG in a visual stimulus-response experiment with 48 participants. We have designed a Fully Connected Neural Network (FCNN) and a Convolutional Neural Network (CNN) to predict and classify RTs for each trial. Though deep neural networks are widely known for classification applications, cascading FCNN/CNN with the Random Forest model, we designed a robust regression-based estimator to predict RT. With the FCNN model, the accuracies obtained for binary and 3-class classification were 93% and 76%, respectively, which further improved with the use of CNN (94% and 78%, respectively). The regression-based approach predicted RTs with correlation coefficients (CC) of 0.78 and 0.80 for FCNN and CNN, respectively. Investigating further, we found that the left central as well as parietal and occipital lobes were crucial for predicting RT, with significant activities in the theta and alpha frequency bands.