More Like this

1. Multidisciplinary Research and Teaching by Means of Employing FTIR Spectroscopic Imaging System and Characterization Techniques

   https://doi.org/10.18260/1-2--34988  

   Alavi, Zahrasadat ( June 2020 , 2020 ASEE Virtual Annual Conference Content Access Proceedings)

   null (Ed.)

   This paper focuses on discussing the efforts made to engage students in multi-disciplinary research and integrate teaching and research in the areas of FTIR Spectro- microscopy and image processing and analysis. The author (PI) and co-PIs acquired a Fourier Transform Infrared (FTIR) Spectroscopic Imaging equipment through the National Science Foundation- Major Research Instrumentation (NSF- MRI) grant (#1827134). This project aims to use the equipment to conduct undergraduate and graduate research projects and teach undergraduate and graduate classes. The NSF awarded the California State University Chico (CSU Chico) $175,305 to acquire an FTIR spectrometer and microscope, which are important tools for chemical characterization of samples with infrared active molecules. FTIR Spectroscopic Imaging System especially provides accurate chemical images that reveal the variations in images’ pixels which are mappings of constituent materials of samples rather than a single visible image with slight variations. By employing this equipment in research and the Image Processing course, students can learn how to collect, process and analyze the imaging data of samples and the corresponding spectral data. The students not only will learn how to process a single chemical image, but also will work with the data cubes to consider the pixel intensities along the IR spectrum, experience working with big data, hone the skills to design experiments, analyze larger data sets, develop pre- and post-image processing techniques, and apply and refine math and programming skills. Image processing course conventionally is based on math, digital signal and systems, and requires programming skills such as Matlab, C++, and Python. along with the mentioned knowledge. Additionally, the research conducted by this equipment promotes collaboration between engineering major students and science major students. In this paper, the author will explain how collecting data through running experiments with the FTIR Spectroscopic Imaging equipment helps students visualize theory and relate it to real world problems. This paper also discusses the results of engaging undergraduate students from various majors in research. Moreover, it will discuss some of the projects that were conducted by undergraduate students and their learning outcomes. The objective of the research projects was material characterization towards contribution to health by employing FTIR Spectroscopic Imaging System. 

   more » « less
2. Enlisting historically excluded undergraduates in the effort to extend knowledge of West Antarctica’s bedrock, through course-based undergraduate research experiences (cures) and Art-Science initiatives

   Siddoway, Christine ; Thomson, Stuart ; Taylor, Jennifer ; Pepper, Marty ; Furlong, Heather ; Ruggiero, Joe ; & Reed, Brandon ( September 2022 , 29th West Antarctic Ice Sheet workshop)

   An enormous reserve of information about the subglacial bedrock, tectonic and topographic evolution of Marie Byrd Land (MBL) exists within glaciomarine sediments of the Amundsen Sea shelf, slope and deep sea, and MBL marine shelf. Investigators of the NSF ICI-Hot and NSF Linchpin projects partnered with Arizona Laserchron Center to provide course-based undergraduate research experiences (CUREs) for from groups who do not ordinarily find access points to Antarctic science. Our courses enlist BIPOC and gender-expansive undergraduates in studies of ice-rafted debris (IRD) and bedrock samples, in order to impart skills, train in the use of research instrumentation, help students to develop confidence in their scientific abilities, and collaboratively address WAIS research questions at an early academic stage. CUREs afford benefits to graduate researchers and postdoctoral scientists, also, who join in as instructional faculty: CUREs allow GRs and PDs to engage in teaching that closely ties to their active research, yet provides practical experience to strengthen the academic portfolio (Cascella & Jez, 2018). Team members also develop art-science initiatives that engage students and community members who may not ordinarily engage with science, forging connections that make science relatable. Re-casting science topics through art centers personal connections and humanizes science, to promote understanding that goes beyond the purely analytical. Academic research shows that diverse undergraduates gain markedly from the convergence of art and science, and from involvement in collaborative research conducted within a CURE cohort, rather than as an individualized experience (e.g. Shanahan et al. 2022). The CUREs are offered as regular courses for credit, making access equitable via course enrollment. The course designation carries a legitimacy that is sought by students who balance academics with part-time employment. Course information is disseminated via STEM Bridge programs and/or an academic advising hub that reaches students from groups that are insufficiently represented within STEM and cryosphere science. CURE investigation of Amundsen Sea and WAIS problems is worthy objective because: 1) A variety of sample preparation, geochemical methods, and scientific best-practices can be imparted, while educating students about Antarctica’s geological configuration and role in the Earth climate system. 2) Individual projects that are narrowly defined can readily scaffold into collaborative science at the time of data synthesis and interpretation. 3) There is a high likelihood of scientific discovery that contributes to grant objectives. 4) Enrolled students will experience ambiguity and instrumentation setbacks alongside their faculty and instructors, and will likely have an opportunity to withstand/overcome challenges in a manner that trains students in complex problem solving and imparts resilience (St John et al., 2019). Based on our experiences, we consider CUREs as a means to create more inclusive and equitable spaces for learning to do research, and a basis for a broadening future WAIS community. Our groups have yet to assess student learning gains and STEM entry in a robust way, but we can report that two presenters at WAIS 2022 came from our 2021 CURE, and four polar science graduate researchers gained experience via CURE teaching. Data obtained by CURE students is contributing to our NSF projects’ aims to obtain isotope, age, and petrogenetic criteria with bearing on the subglacial bedrock geology, tectonic and landscape evolution, and ice sheet history of MBL. Cited and recommended works: Cascella & Jez, 2018, doi: 10.1021/acs.jchemed.7b00705 Gentile et al., 2017, doi: 10.17226/24622 Shanahan et al. 2022, https://www.cur.org/assets/1/23/01-01_TOC_SPUR_Winter21.pdf Shortlidge & Brownell, 2016, doi: 10.1128/jmbe.v17i3.1103 St. John et al. 2019, EOS, doi: 10.1029/2019EO127285. 

   more » « less
3. An innovative active learning module on snow and climate modeling

   https://doi.org/10.3389/frwa.2022.912776  

   Roundy, Joshua K. ; Gallagher, Melissa A. ; Byrd, Jenny L. ( July 2022 , Frontiers in Water)

   The interaction between climate and the hydrologic cycle is complex due to intricate feedback mechanisms that can have multiple impacts on key hydrologic variables. Under a changing climate, it is becoming increasingly important for undergraduate engineering students to have a better understanding of climate and the hydrologic cycle to ensure future engineering systems are more climate resilient. One way of teaching undergraduate students about these key interactions between climate and the hydrologic cycle is through numerical models that mimic these relationships. However, this is difficult to do in an undergraduate engineering course because these models are complex, and it is not feasible to devote class time and resources to teaching students the knowledge base required to run and analyze these numerical models. In addition, the recent COVID-19 pandemic required a rapid change to flexible teaching methods that can be implemented in online, hybrid, or in-person courses. To overcome these limitations, a backward design and constructive alignment approach was used to develop an active learning module in the HydroLearn framework that allows students to explore the connection between snow processes and streamflow and how this will change under different climate scenarios using numerical models and analysis. This learning module provides learning activities and tools that help the student develop a basic knowledge of snow formation and terminology, snow measurements, numerical models of snow processes, and changes in snow and streamflow under future climate. This module is particularly innovative in that it uses Google Colabs and an interactive user interface to facilitate the students' active learning in an environment that is accessible for all students and is sustainable for continued use and adaptation. This paper describes the approach, best practices and lessons learned in developing and implementing this active learning module in a remote and in-person course. In addition, it presents the results from motivation and student self-assessment surveys and discusses opportunities for improvement and further implementation that have implications for the future of hydrologic education. 

   more » « less
4. A Deep Learning-Based Real-time Seizure Detection System

   https://doi.org/10.1109/SPMB50085.2020.9353623  

   Shawki, N. ; Elseify, T. ; Cap, T. ; Shah, V. ; Obeid, I. ; Picone, J. ( December 2020 , Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium (SPMB))

   Obeid, Iyad Selesnick (Ed.)

   Electroencephalography (EEG) is a popular clinical monitoring tool used for diagnosing brain-related disorders such as epilepsy [1]. As monitoring EEGs in a critical-care setting is an expensive and tedious task, there is a great interest in developing real-time EEG monitoring tools to improve patient care quality and efficiency [2]. However, clinicians require automatic seizure detection tools that provide decisions with at least 75% sensitivity and less than 1 false alarm (FA) per 24 hours [3]. Some commercial tools recently claim to reach such performance levels, including the Olympic Brainz Monitor [4] and Persyst 14 [5]. In this abstract, we describe our efforts to transform a high-performance offline seizure detection system [3] into a low latency real-time or online seizure detection system. An overview of the system is shown in Figure 1. The main difference between an online versus offline system is that an online system should always be causal and has minimum latency which is often defined by domain experts. The offline system, shown in Figure 2, uses two phases of deep learning models with postprocessing [3]. The channel-based long short term memory (LSTM) model (Phase 1 or P1) processes linear frequency cepstral coefficients (LFCC) [6] features from each EEG channel separately. We use the hypotheses generated by the P1 model and create additional features that carry information about the detected events and their confidence. The P2 model uses these additional features and the LFCC features to learn the temporal and spatial aspects of the EEG signals using a hybrid convolutional neural network (CNN) and LSTM model. Finally, Phase 3 aggregates the results from both P1 and P2 before applying a final postprocessing step. The online system implements Phase 1 by taking advantage of the Linux piping mechanism, multithreading techniques, and multi-core processors. To convert Phase 1 into an online system, we divide the system into five major modules: signal preprocessor, feature extractor, event decoder, postprocessor, and visualizer. The system reads 0.1-second frames from each EEG channel and sends them to the feature extractor and the visualizer. The feature extractor generates LFCC features in real time from the streaming EEG signal. Next, the system computes seizure and background probabilities using a channel-based LSTM model and applies a postprocessor to aggregate the detected events across channels. The system then displays the EEG signal and the decisions simultaneously using a visualization module. The online system uses C++, Python, TensorFlow, and PyQtGraph in its implementation. The online system accepts streamed EEG data sampled at 250 Hz as input. The system begins processing the EEG signal by applying a TCP montage [8]. Depending on the type of the montage, the EEG signal can have either 22 or 20 channels. To enable the online operation, we send 0.1-second (25 samples) length frames from each channel of the streamed EEG signal to the feature extractor and the visualizer. Feature extraction is performed sequentially on each channel. The signal preprocessor writes the sample frames into two streams to facilitate these modules. In the first stream, the feature extractor receives the signals using stdin. In parallel, as a second stream, the visualizer shares a user-defined file with the signal preprocessor. This user-defined file holds raw signal information as a buffer for the visualizer. The signal preprocessor writes into the file while the visualizer reads from it. Reading and writing into the same file poses a challenge. The visualizer can start reading while the signal preprocessor is writing into it. To resolve this issue, we utilize a file locking mechanism in the signal preprocessor and visualizer. Each of the processes temporarily locks the file, performs its operation, releases the lock, and tries to obtain the lock after a waiting period. The file locking mechanism ensures that only one process can access the file by prohibiting other processes from reading or writing while one process is modifying the file [9]. The feature extractor uses circular buffers to save 0.3 seconds or 75 samples from each channel for extracting 0.2-second or 50-sample long center-aligned windows. The module generates 8 absolute LFCC features where the zeroth cepstral coefficient is replaced by a temporal domain energy term. For extracting the rest of the features, three pipelines are used. The differential energy feature is calculated in a 0.9-second absolute feature window with a frame size of 0.1 seconds. The difference between the maximum and minimum temporal energy terms is calculated in this range. Then, the first derivative or the delta features are calculated using another 0.9-second window. Finally, the second derivative or delta-delta features are calculated using a 0.3-second window [6]. The differential energy for the delta-delta features is not included. In total, we extract 26 features from the raw sample windows which add 1.1 seconds of delay to the system. We used the Temple University Hospital Seizure Database (TUSZ) v1.2.1 for developing the online system [10]. The statistics for this dataset are shown in Table 1. A channel-based LSTM model was trained using the features derived from the train set using the online feature extractor module. A window-based normalization technique was applied to those features. In the offline model, we scale features by normalizing using the maximum absolute value of a channel [11] before applying a sliding window approach. Since the online system has access to a limited amount of data, we normalize based on the observed window. The model uses the feature vectors with a frame size of 1 second and a window size of 7 seconds. We evaluated the model using the offline P1 postprocessor to determine the efficacy of the delayed features and the window-based normalization technique. As shown by the results of experiments 1 and 4 in Table 2, these changes give us a comparable performance to the offline model. The online event decoder module utilizes this trained model for computing probabilities for the seizure and background classes. These posteriors are then postprocessed to remove spurious detections. The online postprocessor receives and saves 8 seconds of class posteriors in a buffer for further processing. It applies multiple heuristic filters (e.g., probability threshold) to make an overall decision by combining events across the channels. These filters evaluate the average confidence, the duration of a seizure, and the channels where the seizures were observed. The postprocessor delivers the label and confidence to the visualizer. The visualizer starts to display the signal as soon as it gets access to the signal file, as shown in Figure 1 using the “Signal File” and “Visualizer” blocks. Once the visualizer receives the label and confidence for the latest epoch from the postprocessor, it overlays the decision and color codes that epoch. The visualizer uses red for seizure with the label SEIZ and green for the background class with the label BCKG. Once the streaming finishes, the system saves three files: a signal file in which the sample frames are saved in the order they were streamed, a time segmented event (TSE) file with the overall decisions and confidences, and a hypotheses (HYP) file that saves the label and confidence for each epoch. The user can plot the signal and decisions using the signal and HYP files with only the visualizer by enabling appropriate options. For comparing the performance of different stages of development, we used the test set of TUSZ v1.2.1 database. It contains 1015 EEG records of varying duration. The any-overlap performance [12] of the overall system shown in Figure 2 is 40.29% sensitivity with 5.77 FAs per 24 hours. For comparison, the previous state-of-the-art model developed on this database performed at 30.71% sensitivity with 6.77 FAs per 24 hours [3]. The individual performances of the deep learning phases are as follows: Phase 1’s (P1) performance is 39.46% sensitivity and 11.62 FAs per 24 hours, and Phase 2 detects seizures with 41.16% sensitivity and 11.69 FAs per 24 hours. We trained an LSTM model with the delayed features and the window-based normalization technique for developing the online system. Using the offline decoder and postprocessor, the model performed at 36.23% sensitivity with 9.52 FAs per 24 hours. The trained model was then evaluated with the online modules. The current performance of the overall online system is 45.80% sensitivity with 28.14 FAs per 24 hours. Table 2 summarizes the performances of these systems. The performance of the online system deviates from the offline P1 model because the online postprocessor fails to combine the events as the seizure probability fluctuates during an event. The modules in the online system add a total of 11.1 seconds of delay for processing each second of the data, as shown in Figure 3. In practice, we also count the time for loading the model and starting the visualizer block. When we consider these facts, the system consumes 15 seconds to display the first hypothesis. The system detects seizure onsets with an average latency of 15 seconds. Implementing an automatic seizure detection model in real time is not trivial. We used a variety of techniques such as the file locking mechanism, multithreading, circular buffers, real-time event decoding, and signal-decision plotting to realize the system. A video demonstrating the system is available at: https://www.isip.piconepress.com/projects/nsf_pfi_tt/resources/videos/realtime_eeg_analysis/v2.5.1/video_2.5.1.mp4. The final conference submission will include a more detailed analysis of the online performance of each module. ACKNOWLEDGMENTS Research reported in this publication was most recently supported by the National Science Foundation Partnership for Innovation award number IIP-1827565 and the Pennsylvania Commonwealth Universal Research Enhancement Program (PA CURE). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the official views of any of these organizations. REFERENCES [1] A. Craik, Y. He, and J. L. Contreras-Vidal, “Deep learning for electroencephalogram (EEG) classification tasks: a review,” J. Neural Eng., vol. 16, no. 3, p. 031001, 2019. https://doi.org/10.1088/1741-2552/ab0ab5. [2] A. C. Bridi, T. Q. Louro, and R. C. L. Da Silva, “Clinical Alarms in intensive care: implications of alarm fatigue for the safety of patients,” Rev. Lat. Am. Enfermagem, vol. 22, no. 6, p. 1034, 2014. https://doi.org/10.1590/0104-1169.3488.2513. [3] M. Golmohammadi, V. Shah, I. Obeid, and J. Picone, “Deep Learning Approaches for Automatic Seizure Detection from Scalp Electroencephalograms,” in Signal Processing in Medicine and Biology: Emerging Trends in Research and Applications, 1st ed., I. Obeid, I. Selesnick, and J. Picone, Eds. New York, New York, USA: Springer, 2020, pp. 233–274. https://doi.org/10.1007/978-3-030-36844-9_8. [4] “CFM Olympic Brainz Monitor.” [Online]. Available: https://newborncare.natus.com/products-services/newborn-care-products/newborn-brain-injury/cfm-olympic-brainz-monitor. [Accessed: 17-Jul-2020]. [5] M. L. Scheuer, S. B. Wilson, A. Antony, G. Ghearing, A. Urban, and A. I. Bagic, “Seizure Detection: Interreader Agreement and Detection Algorithm Assessments Using a Large Dataset,” J. Clin. Neurophysiol., 2020. https://doi.org/10.1097/WNP.0000000000000709. [6] A. Harati, M. Golmohammadi, S. Lopez, I. Obeid, and J. Picone, “Improved EEG Event Classification Using Differential Energy,” in Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium, 2015, pp. 1–4. https://doi.org/10.1109/SPMB.2015.7405421. [7] V. Shah, C. Campbell, I. Obeid, and J. Picone, “Improved Spatio-Temporal Modeling in Automated Seizure Detection using Channel-Dependent Posteriors,” Neurocomputing, 2021. [8] W. Tatum, A. Husain, S. Benbadis, and P. Kaplan, Handbook of EEG Interpretation. New York City, New York, USA: Demos Medical Publishing, 2007. [9] D. P. Bovet and C. Marco, Understanding the Linux Kernel, 3rd ed. O’Reilly Media, Inc., 2005. https://www.oreilly.com/library/view/understanding-the-linux/0596005652/. [10] V. Shah et al., “The Temple University Hospital Seizure Detection Corpus,” Front. Neuroinform., vol. 12, pp. 1–6, 2018. https://doi.org/10.3389/fninf.2018.00083. [11] F. Pedregosa et al., “Scikit-learn: Machine Learning in Python,” J. Mach. Learn. Res., vol. 12, pp. 2825–2830, 2011. https://dl.acm.org/doi/10.5555/1953048.2078195. [12] J. Gotman, D. Flanagan, J. Zhang, and B. Rosenblatt, “Automatic seizure detection in the newborn: Methods and initial evaluation,” Electroencephalogr. Clin. Neurophysiol., vol. 103, no. 3, pp. 356–362, 1997. https://doi.org/10.1016/S0013-4694(97)00003-9. 

   more » « less
5. An On-Demand Service for Managing and Analyzing Arctic Sea Ice High Spatial Resolution Imagery

   https://doi.org/10.3390/data5020039  

   Sha, Dexuan ; Miao, Xin ; Xu, Mengchao ; Yang, Chaowei ; Xie, Hongjie ; Mestas-Nuñez, Alberto M. ; Li, Yun ; Liu, Qian ; Yang, Jingchao ( June 2020 , Data)

   null (Ed.)

   Sea ice acts as both an indicator and an amplifier of climate change. High spatial resolution (HSR) imagery is an important data source in Arctic sea ice research for extracting sea ice physical parameters, and calibrating/validating climate models. HSR images are difficult to process and manage due to their large data volume, heterogeneous data sources, and complex spatiotemporal distributions. In this paper, an Arctic Cyberinfrastructure (ArcCI) module is developed that allows a reliable and efficient on-demand image batch processing on the web. For this module, available associated datasets are collected and presented through an open data portal. The ArcCI module offers an architecture based on cloud computing and big data components for HSR sea ice images, including functionalities of (1) data acquisition through File Transfer Protocol (FTP) transfer, front-end uploading, and physical transfer; (2) data storage based on Hadoop distributed file system and matured operational relational database; (3) distributed image processing including object-based image classification and parameter extraction of sea ice features; (4) 3D visualization of dynamic spatiotemporal distribution of extracted parameters with flexible statistical charts. Arctic researchers can search and find arctic sea ice HSR image and relevant metadata in the open data portal, obtain extracted ice parameters, and conduct visual analytics interactively. Users with large number of images can leverage the service to process their image in high performance manner on cloud, and manage, analyze results in one place. The ArcCI module will assist domain scientists on investigating polar sea ice, and can be easily transferred to other HSR image processing research projects. 

   more » « less