More Like this

1. Gold Nanoparticle-Modified Carbon-Fiber Microelectrodes for the Electrochemical Detection of Cd2+ via Fast-Scan Cyclic Voltammetry

   https://doi.org/10.3390/mi15030294  

   Manring, Noel ; Strini, Miriam ; Koifman, Gene ; Smeltz, Jessica L. ; Pathirathna, Pavithra ( March 2024 , Micromachines)

   Neurotoxic heavy metals, such as Cd2+, pose a significant global health concern due to their increased environmental contamination and subsequent detrimental health hazards they pose to human beings. These metal ions can breach the blood-brain barrierblood–brain barrier, leading to severe and often irreversible damage to the central nervous system and other vital organs. Therefore, developing a highly sensitive, robust, and rapid in vivo detection method for these hazardous heavy metal ions is of the utmost importance for early detection, thus initiating timely therapeutics. Detecting ultra-low levels of toxic metal ions in vivo and obtaining accurate speciation information remains a challenge with conventional analytical techniques. In this study, we fabricated a novel carbon carbon-fiber microelectrode (CFM)-based sensor that can detect Cd2+ ions using fast-scan cyclic voltammetry by electrodepositing gold nanoparticles (AuNP). We optimized electrochemical parameters that generate a unique cyclic voltammogram (CV) of Cd2+ at a temporal resolution of 100 ms with our novel sensor. All our experiments were performed in tris buffer that mimics the artificial cerebellum fluid. We established a calibration curve resulting in a limit of detection (LOD) of 0.01 µM with a corresponding sensitivity of 418.02 nA/ µM. The sensor’s selectivity was evaluated in the presence of other metal ions, and it was noteworthy to observe that the sensor retained its ability to produce the distinctive Cd2+ CV, even when the concentration of other metal ions was 200 times higher than that of Cd2+. We also found that our sensor could detect free Cd2+ ions in the presence of complexing agents. Furthermore, we analyzed the solution chemistry of each of those Cd2+–ligand solutions using a geochemical model, PHREEQC. The concentrations of free Cd2+ ions determined through our electrochemical data align well with geochemical modeling data, thus validating the response of our novel sensor. Furthermore, we reassessed our sensor’s LOD in tris buffer based on the concentration of free Cd2+ ions determined through PHREEQC analysis, revealing an LOD of 0.00132 µM. We also demonstrated the capability of our sensor to detect Cd2+ ions in artificial urine samples, showcasing its potential for application in actual biological samples. To the best of our knowledge, this is the first AuNP-modified, CFM-based Cd2+ sensor capable of detecting ultra-low concentrations of free Cd2+ ions in different complex matrices, including artificial urine at a temporal resolution of 100 ms, making it an excellent analytical tool for future real-time, in vivo detection, particularly in the brain.

    

   more » « less
2. 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
3. Employing Conductive Metal-Organic Frameworks for Voltammetric Detection of Neurochemicals

   https://doi.org/10.1021/jacs.9b13402  

   Ko, Michael ; Mendecki, Lukasz ; Eagleton, Aileen M. ; Durbin, Claudia G. ; Stolz, Robert M. ; Meng, Zheng ; Mirica, Katherine A ( March 2020 , Journal of the American Chemical Society)

   This paper describes the first implementation of an array of two-dimensional (2D) layered conductive metal–organic frameworks (MOFs) as drop-casted film electrodes that facilitate voltammetric detection of redox active neurochemicals in a multi-analyte solution. The device configuration comprises a glassy carbon electrode modified with a film of conductive MOF (M3HXTP2; M = Ni, Cu; and X = NH, 2,3,6,7,10,11-hexaiminotriphenylene (HITP) or O, 2,3,6,7,10,11-hexahydroxytriphenylene (HHTP)). The utility of 2D MOFs in voltammetric sensing is measured by the detection of ascorbic acid (AA), dopamine (DA), uric acid (UA), and serotonin (5-HT) in 0.1 M PBS (pH=7.4). In particular, Ni3HHTP2 MOF demonstrated nanomolar detection limits of 63 ± 11 nM for DA and 40 ± 17 nM for 5-HT through a wide concentration range (40 nM – 200 µM). The applicability in biologically-relevant detection was further demonstrated in simulated urine using Ni3HHTP2 MOFs for the detection of 5-HT with nanomolar detection limit of 63 ± 11 nM for 5-HT through a wide concentration range (63 nM – 200 µM) in the presence of constant background of DA. The implementation of conductive MOFs in voltammetric detection holds promise for further development of highly modular, sensitive, selective, and stable electroanalytical devices. 

   more » « less
4. Highly porous gold supraparticles as surface-enhanced Raman spectroscopy (SERS) substrates for sensitive detection of environmental contaminants

   https://doi.org/10.1039/d2ra06248h  

   Kang, Seju ; Wang, Wei ; Rahman, Asifur ; Nam, Wonil ; Zhou, Wei ; Vikesland, Peter J. ( November 2022 , RSC Advances)

   Surface-enhanced Raman spectroscopy (SERS) has great potential as an analytical technique for environmental analyses. In this study, we fabricated highly porous gold (Au) supraparticles ( i.e. , ∼100 μm diameter agglomerates of primary nano-sized particles) and evaluated their applicability as SERS substrates for the sensitive detection of environmental contaminants. Facile supraparticle fabrication was achieved by evaporating a droplet containing an Au and polystyrene (PS) nanoparticle mixture on a superamphiphobic nanofilament substrate. Porous Au supraparticles were obtained through the removal of the PS phase by calcination at 500 °C. The porosity of the Au supraparticles was readily adjusted by varying the volumetric ratios of Au and PS nanoparticles. Six environmental contaminants (malachite green isothiocyanate, rhodamine B, benzenethiol, atrazine, adenine, and gene segment) were successfully adsorbed to the porous Au supraparticles, and their distinct SERS spectra were obtained. The observed linear dependence of the characteristic Raman peak intensity for each environmental contaminant on its aqueous concentration reveals the quantitative SERS detection capability by porous Au supraparticles. The limit of detection (LOD) for the six environmental contaminants ranged from ∼10 nM to ∼10 μM, which depends on analyte affinity to the porous Au supraparticles and analyte intrinsic Raman cross-sections. The porous Au supraparticles enabled multiplex SERS detection and maintained comparable SERS detection sensitivity in wastewater influent. Overall, we envision that the Au supraparticles can potentially serve as practical and sensitive SERS devices for environmental analysis applications. 

   more » « less
5. A portable electrochemical sensing platform for serotonin detection based on surface-modified carbon fiber microelectrodes

   https://doi.org/10.1039/D2AY01627C  

   Han, Jinjing ; Stine, Justin M. ; Chapin, Ashley A. ; Ghodssi, Reza ( March 2023 , Analytical Methods)

   Serotonin (5-HT) is one of the key neurotransmitters in the human body, regulating numerous physiological functions. A disruption in 5-HT homeostasis could result in serious health problems, including neurodegenerative disorders, depression, and 5-HT syndrome. Detection of 5-HT concentrations in biological fluids, such as urine, is a potential solution for early diagnosis of these diseases. In this study, we developed a novel, simple, and low-cost electrochemical sensing platform consisting of a portable workstation with customized electrodes for 5-HT detection in artificial biological fluids. Nafion/carbon nanotubes (CNTs) and electrochemically modified carbon fiber microelectrodes (Nafion–CNT/EC CFMEs) displayed improved 5-HT sensitivity and selectivity. Together with a customized Ag/AgCl reference electrode and Pt counter electrode, the portable 5-HT sensing platform had a sensitivity of 0.074 μA μM −1 and a limit of detection (LOD) of 140 nM. This system was also assessed to measure 5-HT spiked in artificial urine samples, showing nearly full recovery rates. These satisfactory results demonstrated that the portable system exhibits outstanding performance and confirmed the feasibility of 5-HT detection, which can be used to provide point-of-care analysis in actual biological samples. 

   more » « less