Artificial intelligence-enhanced epileptic seizure detection by wearables.

OBJECTIVE: Wrist- or ankle-worn devices are less intrusive than the widely used electroencephalographic (EEG) systems for monitoring epileptic seizures. Using custom-developed deep-learning seizure detection models, we demonstrate the detection of a broad range of seizure types by wearable signals. METHODS: Patients admitted to the epilepsy monitoring unit were enrolled and asked to wear wearable sensors on either wrists or ankles. We collected patients' electrodermal activity, accelerometry (ACC), and photoplethysmography, from which blood volume pulse (BVP) is derived. Board-certified epileptologists determined seizure onset, offset, and types using video and EEG recordings per the International League Against Epilepsy 2017 classification. We applied three neural network models-a convolutional neural network (CNN) and a CNN-long short-term memory (LSTM)-based generalized detection model and an autoencoder-based personalized detection model-to the raw time-series sensor data to detect seizures and utilized performance measures, including sensitivity, false positive rate (the number of false alarms divided by the total number of nonseizure segments), number of false alarms per day, and detection delay. We applied a 10-fold patientwise cross-validation scheme to the multisignal biosensor data and evaluated model performance on 28 seizure types. RESULTS: We analyzed 166 patients (47.6% female, median age = 10.0 years) and 900 seizures (13 254 h of sensor data) for 28 seizure types. With a CNN-LSTM-based seizure detection model, ACC, BVP, and their fusion performed better than chance; ACC and BVP data fusion reached the best detection performance of 83.9% sensitivity and 35.3% false positive rate. Nineteen of 28 seizure types could be detected by at least one data modality with area under receiver operating characteristic curve > .8 performance. SIGNIFICANCE: Results from this in-hospital study contribute to a paradigm shift in epilepsy care that entails noninvasive seizure detection, provides time-sensitive and accurate data on additional clinical seizure types, and proposes a novel combination of an out-of-the-box monitoring algorithm with an individualized person-oriented seizure detection approach.

Attention is not all you need: the complicated case of ethically using large language models in healthcare and medicine.

Large Language Models (LLMs) are a key component of generative artificial intelligence (AI) applications for creating new content including text, imagery, audio, code, and videos in response to textual instructions. Without human oversight, guidance and responsible design and operation, such generative AI applications will remain a party trick with substantial potential for creating and spreading misinformation or harmful and inaccurate content at unprecedented scale. However, if positioned and developed responsibly as companions to humans augmenting but not replacing their role in decision making, knowledge retrieval and other cognitive processes, they could evolve into highly efficient, trustworthy, assistive tools for information management. This perspective describes how such tools could transform data management workflows in healthcare and medicine, explains how the underlying technology works, provides an assessment of risks and limitations, and proposes an ethical, technical, and cultural framework for responsible design, development, and deployment. It seeks to incentivise users, developers, providers, and regulators of generative AI that utilises LLMs to collectively prepare for the transformational role this technology could play in evidence-based sectors.

Features importance in seizure classification using scalp EEG reduced to single timeseries.

Seizure detection and seizure-type classification are best performed using intra-cranial or full-scalp electroencephalogram (EEG). In embedded wearable systems however, recordings from only a few electrodes are available, reducing the spatial resolution of the signals to a handful of timeseries at most. Taking this constraint into account, we tested the performance of multiple classifiers using a subset of the EEG recordings by selecting a single trace from the montage or performing a dimensionality reduction over each hemispherical space. Our results support that Random Forest (RF) classifiers lead most efficient and stable classification performances over Support Vector Machines (SVM). Interestingly, tracking the feature importances using permutation tests reveals that classical EEG spectrum power bands display different rankings across the classifiers: low frequencies (delta, theta) are most important for SVMs while higher frequencies (alpha, gamma) are more relevant for RF and Decision Trees. We reach up to 94.3% ∓ 5.3% accuracy in classifying absence from tonic-clonic seizures using state-of-art sampling methods for unbalanced datasets and leave-patients-out 3-fold cross-validation policy.

Seizure detection using wearable sensors and machine learning: Setting a benchmark.

OBJECTIVE: Tracking seizures is crucial for epilepsy monitoring and treatment evaluation. Current epilepsy care relies on caretaker seizure diaries, but clinical seizure monitoring may miss seizures. Wearable devices may be better tolerated and more suitable for long-term ambulatory monitoring. This study evaluates the seizure detection performance of custom-developed machine learning (ML) algorithms across a broad spectrum of epileptic seizures utilizing wrist- and ankle-worn multisignal biosensors. METHODS: We enrolled patients admitted to the epilepsy monitoring unit and asked them to wear a wearable sensor on either their wrists or ankles. The sensor recorded body temperature, electrodermal activity, accelerometry (ACC), and photoplethysmography, which provides blood volume pulse (BVP). We used electroencephalographic seizure onset and offset as determined by a board-certified epileptologist as a standard comparison. We trained and validated ML for two different algorithms: Algorithm 1, ML methods for developing seizure type-specific detection models for nine individual seizure types; and Algorithm 2, ML methods for building general seizure type-agnostic detection, lumping together all seizure types. RESULTS: We included 94 patients (57.4% female, median age = 9.9 years) and 548 epileptic seizures (11 066 h of sensor data) for a total of 930 seizures and nine seizure types. Algorithm 1 detected eight of nine seizure types better than chance (area under the receiver operating characteristic curve [AUC-ROC] = .648-.976). Algorithm 2 detected all nine seizure types better than chance (AUC-ROC = .642-.995); a fusion of ACC and BVP modalities achieved the best AUC-ROC (.752) when combining all seizure types together. SIGNIFICANCE: Automatic seizure detection using ML from multimodal wearable sensor data is feasible across a broad spectrum of epileptic seizures. Preliminary results show better than chance seizure detection. The next steps include validation of our results in larger datasets, evaluation of the detection utility tool for additional clinical seizure types, and integration of additional clinical information.

Evaluation of artificial intelligence systems for assisting neurologists with fast and accurate annotations of scalp electroencephalography data.

BACKGROUND: Assistive automatic seizure detection can empower human annotators to shorten patient monitoring data review times. We present a proof-of-concept for a seizure detection system that is sensitive, automated, patient-specific, and tunable to maximise sensitivity while minimizing human annotation times. The system uses custom data preparation methods, deep learning analytics and electroencephalography (EEG) data. METHODS: Scalp EEG data of 365 patients containing 171,745 s ictal and 2,185,864 s interictal samples obtained from clinical monitoring systems were analysed as part of a crowdsourced artificial intelligence (AI) challenge. Participants were tasked to develop an ictal/interictal classifier with high sensitivity and low false alarm rates. We built a challenge platform that prevented participants from downloading or directly accessing the data while allowing crowdsourced model development. FINDINGS: The automatic detection system achieved tunable sensitivities between 75.00% and 91.60% allowing a reduction in the amount of raw EEG data to be reviewed by a human annotator by factors between 142x, and 22x respectively. The algorithm enables instantaneous reviewer-managed optimization of the balance between sensitivity and the amount of raw EEG data to be reviewed. INTERPRETATION: This study demonstrates the utility of deep learning for patient-specific seizure detection in EEG data. Furthermore, deep learning in combination with a human reviewer can provide the basis for an assistive data labelling system lowering the time of manual review while maintaining human expert annotation performance. FUNDING: IBM employed all IBM Research authors. Temple University employed all Temple University authors. The Icahn School of Medicine at Mount Sinai employed Eren Ahsen. The corresponding authors Stefan Harrer and Gustavo Stolovitzky declare that they had full access to all the data in the study and that they had final responsibility for the decision to submit for publication.

Evaluation of Combined Artificial Intelligence and Radiologist Assessment to Interpret Screening Mammograms.

Importance: Mammography screening currently relies on subjective human interpretation. Artificial intelligence (AI) advances could be used to increase mammography screening accuracy by reducing missed cancers and false positives. Objective: To evaluate whether AI can overcome human mammography interpretation limitations with a rigorous, unbiased evaluation of machine learning algorithms. Design, Setting, and Participants: In this diagnostic accuracy study conducted between September 2016 and November 2017, an international, crowdsourced challenge was hosted to foster AI algorithm development focused on interpreting screening mammography. More than 1100 participants comprising 126 teams from 44 countries participated. Analysis began November 18, 2016. Main Outcomes and Measurements: Algorithms used images alone (challenge 1) or combined images, previous examinations (if available), and clinical and demographic risk factor data (challenge 2) and output a score that translated to cancer yes/no within 12 months. Algorithm accuracy for breast cancer detection was evaluated using area under the curve and algorithm specificity compared with radiologists' specificity with radiologists' sensitivity set at 85.9% (United States) and 83.9% (Sweden). An ensemble method aggregating top-performing AI algorithms and radiologists' recall assessment was developed and evaluated. Results: Overall, 144 231 screening mammograms from 85 580 US women (952 cancer positive ≤12 months from screening) were used for algorithm training and validation. A second independent validation cohort included 166 578 examinations from 68 008 Swedish women (780 cancer positive). The top-performing algorithm achieved an area under the curve of 0.858 (United States) and 0.903 (Sweden) and 66.2% (United States) and 81.2% (Sweden) specificity at the radiologists' sensitivity, lower than community-practice radiologists' specificity of 90.5% (United States) and 98.5% (Sweden). Combining top-performing algorithms and US radiologist assessments resulted in a higher area under the curve of 0.942 and achieved a significantly improved specificity (92.0%) at the same sensitivity. Conclusions and Relevance: While no single AI algorithm outperformed radiologists, an ensemble of AI algorithms combined with radiologist assessment in a single-reader screening environment improved overall accuracy. This study underscores the potential of using machine learning methods for enhancing mammography screening interpretation.

Artificial Intelligence for Clinical Trial Design.

Clinical trials consume the latter half of the 10 to 15 year, 1.5-2.0 billion USD, development cycle for bringing a single new drug to market. Hence, a failed trial sinks not only the investment into the trial itself but also the preclinical development costs, rendering the loss per failed clinical trial at 800 million to 1.4 billion USD. Suboptimal patient cohort selection and recruiting techniques, paired with the inability to monitor patients effectively during trials, are two of the main causes for high trial failure rates: only one of 10 compounds entering a clinical trial reaches the market. We explain how recent advances in artificial intelligence (AI) can be used to reshape key steps of clinical trial design towards increasing trial success rates.

A Robust Low-Cost EEG Motor Imagery-Based Brain-Computer Interface.

Motor imagery (MI) based Brain-Computer Interfaces (BCIs) are a viable option for giving locked-in syndrome patients independence and communicability. BCIs comprising expensive medical-grade EEG systems evaluated in carefully-controlled, artificial environments are impractical for take-home use. Previous studies evaluated low-cost systems; however, performance was suboptimal or inconclusive. Here we evaluated a low-cost EEG system, OpenBCI, in a natural environment and leveraged neurofeedback, deep learning, and wider temporal windows to improve performance. $\mu-$rhythm data collected over the sensorimotor cortex from healthy participants performing relaxation and right-handed MI tasks were used to train a multi-layer perceptron binary classifier using deep learning. We showed that our method outperforms previous OpenBCI MI-based BCIs, thereby extending the BCI capabilities of this low-cost device.

Deep Learning Enabled Automatic Abnormal EEG Identification.

In hospitals, physicians diagnose brain-related disorders such as epilepsy by analyzing electroencephalograms (EEG). However, manual analysis of EEG data requires highly trained clinicians or neurophysiologists and is a procedure that is known to have relatively low inter-rater agreement (IRA). Moreover, the volume of the data and rate at which new data is acquired makes interpretation a time-consuming, resource hungry, and expensive process. In contrast, automated analysis offers the potential to improve the quality of patient care by shortening the time to diagnosis, reducing manual error, and automatically detecting debilitating events. In this paper, we focus on one of the early decisions made in this process which is identifying whether an EEG session is normal or abnormal. Unlike previous approaches, we do not extract hand-engineered features but employ deep neural networks that automatically learn meaningful representations. We undertake a holistic study by exploring various pre-processing techniques and machine learning algorithms for addressing this problem and compare their performance. We have used the recently released "TUH Abnormal EEG Corpus" dataset for evaluating the performance of these algorithms. We show that modern deep gated recurrent neural networks achieve 3.47% better performance than previously reported results.

Epileptic Seizure Prediction Using Big Data and Deep Learning: Toward a Mobile System.

BACKGROUND: Seizure prediction can increase independence and allow preventative treatment for patients with epilepsy. We present a proof-of-concept for a seizure prediction system that is accurate, fully automated, patient-specific, and tunable to an individual's needs. METHODS: Intracranial electroencephalography (iEEG) data of ten patients obtained from a seizure advisory system were analyzed as part of a pseudoprospective seizure prediction study. First, a deep learning classifier was trained to distinguish between preictal and interictal signals. Second, classifier performance was tested on held-out iEEG data from all patients and benchmarked against the performance of a random predictor. Third, the prediction system was tuned so sensitivity or time in warning could be prioritized by the patient. Finally, a demonstration of the feasibility of deployment of the prediction system onto an ultra-low power neuromorphic chip for autonomous operation on a wearable device is provided. RESULTS: The prediction system achieved mean sensitivity of 69% and mean time in warning of 27%, significantly surpassing an equivalent random predictor for all patients by 42%. CONCLUSION: This study demonstrates that deep learning in combination with neuromorphic hardware can provide the basis for a wearable, real-time, always-on, patient-specific seizure warning system with low power consumption and reliable long-term performance.

TrueNorth-enabled real-time classification of EEG data for brain-computer interfacing.

Brain-computer interfaces are commonly proposed to assist individuals with locked-in syndrome to interact with the world around them. In this paper, we present a pipeline to move from recorded brain signals to real-time classification on a low-power platform, such as IBM's TrueNorth Neurosynaptic System. Our results on a EEG-based hand squeeze task show that using a convolutional neural network and a time preserving signal representation strategy provides a good balance between high accuracy and feasibility in a real-time application. This pathway can be adapted to the management of a variety of conditions, including spinal cord injury, epilepsy and Parkinson's disease.

Geometric dependence of the conductance drop in a nanopore due to a particle.

We study the effect of a neutral particle on the ionic flow through a nanopore using a basic uniform field theory and the coupled Poisson-Nernst-Planck and Navier-Stokes (PNP-NS) equations. We consider hourglass and cylindrical pore profiles and examine how the difference in pore shape changes the position dependence of the current change due to the particle. Good quantitative agreement between both calculations is seen, though we find that the simple theory is unable to correctly capture the change in the access resistance of the pore if a particle is placed at the pore entrance. Finally, we examine the spatial variations in the solutions of the PNP-NS equations, finding that the electro-osmotic flow through the pore is completely disrupted for sufficiently large particles.

Sensing of protein molecules through nanopores: a molecular dynamics study.

Solid-state nanopores have been shown to be suitable for single molecule detection. While numerous modeling investigations exist for DNA within nanopores, there are few simulations of protein translocations. In this paper, we use atomistic molecular dynamics to investigate the translocation of proteins through a silicon nitride nanopore. The nanopore dimensions and profile are representative of experimental systems. We are able to calculate the change in blockade current and friction coefficient for different positions of the protein within the pore. The change in ionic current is found to be negligible until the protein is fully within the pore and the current is lowest when the protein is in the pore center. Using a simple theory that gives good quantitative agreement with the simulation results we are able to show that the variation in current with position is a function of the pore shape. In simulations that guide the protein through the nanopore we identify the effect that confinement has on the friction coefficient of the protein. This integrated view of translocation at the nanoscale provides useful insights that can be used to guide the design of future devices.

Regulating the transport of DNA through biofriendly nanochannels in a thin solid membrane.

Channels formed by membrane proteins regulate the transport of water, ions or nutrients that are essential to cells' metabolism. Recent advances in nanotechnology allow us to fabricate solid-state nanopores for transporting and analyzing biomolecules. However, uncontrollable surface properties of a fabricated nanopore cause irregular transport of biomolecules, limiting potential biomimetic applications. Here we show that a nanopore functionalized with a self-assembled monolayer (SAM) can potentially regulate the transport of a DNA molecule by changing functional groups of the SAM. We found that an enhanced interaction between DNA and a SAM-coated nanopore can slow down the translocation speed of DNA molecules and increase the DNA capture-rate. Our results demonstrate that the transport of DNA molecules inside nanopores could be modulated by coating a SAM on the pore surface. Our method to control the DNA motion inside a nanopore may find its applications in nanopore-based DNA sequencing devices.

Dynamics of DNA translocation in a solid-state nanopore immersed in aqueous glycerol.

Nanopore-based technologies have attracted much attention recently for their promising use in low-cost and high-throughput genome sequencing. To achieve single-base resolution of DNA sequencing, it is critical to slow and control the translocation of DNA, which has been achieved in a protein nanopore but not yet in a solid-state nanopore. Using all-atom molecular dynamics simulations, we investigated the dynamics of a single-stranded DNA (ssDNA) molecule in an aqueous glycerol solution confined in a SiO(2) nanopore. The friction coefficient ξ of the ssDNA molecule is found to be approximately 18 times larger in glycerol than in water, which can dramatically slow the motion of ssDNA. The electrophoretic mobility µ of ssDNA in glycerol, however, decreases by almost the same factor, yielding the effective charge (ξµ) of ssDNA being roughly the same as in water. This is counterintuitive since the ssDNA effective charge predicted from the counterion condensation theory varies with the dielectric constant of a solvent. Due to the larger friction coefficient of ssDNA in glycerol, we further show that glycerol can improve trapping of ssDNA in the DNA transistor, a nanodevice that can be used to control the motion of ssDNA in a solid-state nanopore. Simulation results of slowing ssDNA translocation were confirmed in our nanopore experiment.

Electrochemical protection of thin film electrodes in solid state nanopores.

Solid state nanopores are a core element of next-generation single molecule tools in the field of nano-biotechnology. Thin film electrodes integrated into a pore can interact with charges and fields within the pore. In order to keep the nanopore open and thus functional electrochemically induced surface alteration of electrode surfaces and bubble formation inside the pore have to be eliminated. This paper provides electrochemical analyses of nanopores drilled into TiN membranes which in turn were employed as thin film electrodes. We studied physical pore integrity and the occurrence of water decomposition yielding bubble formation inside pores by applying voltages between -4.5 and +4.5 V to membranes in various protection stages continuously for up to 24 h. During potential application pores were exposed to selected electrolyte-solvent systems. We have investigated and successfully eliminated electrochemical pore oxidation and reduction as well as water decomposition inside nanopores of various diameters ranging from 3.5 to 25 nm in 50 nm thick TiN membranes by passivating the nanopores with a plasma-oxidized layer and using a 90% solution of glycerol in water as KCl solvent. Nanopore ionic conductances were measured before and after voltage application in order to test for changes in pore diameter due to electrochemical oxidation or reduction. TEM imaging was used to confirm these observations. While non-passivated pores were electrochemically oxidized, neither electrochemical oxidation nor reduction was observed for passivated pores. Bubble formation through water decomposition could be detected in non-passivated pores in KCl/water solutions but was not observed in 90% glycerol solutions. The use of a protective self-assembled monolayer of hexadecylphosphonic acid (HDPA) was also investigated.

Tribological effects on DNA translocation in a nanochannel coated with a self-assembled monolayer.

A biomimetic nanochannel coated with a self-assembled monolayer (SAM) can be used for sensing and analyzing biomolecules. The interaction between a transported biomolecule and a SAM governs the mechanically or electrically driven motion of the molecule. To investigate the translocation dynamics of a biomolecule, we performed all-atom molecular dynamics simulations on a single-stranded DNA in a solid-state nanochannel coated with a SAM that consists of octane or octanol polymers. Simulation results demonstrate that the interaction between DNA and a hydrophobic or a hydrophilic SAM is effectively repulsive or adhesive, respectively, resulting in different translocation dynamics of DNA. Therefore, with proper designs of SAMs coated on a channel surface, it is possible to control the translocation dynamics of a biomolecule. This work also demonstrates that traditional tribology methods can be deployed to study a biological or biomimetic transport process.

Electrochemical characterization of thin film electrodes toward developing a DNA transistor.

The DNA-Transistor is a device designed to control the translocation of single-stranded DNA through a solid-state nanopore. Functionality of the device is enabled by three electrodes exposed to the DNA-containing electrolyte solution within the pore and the application of a dynamic electrostatic potential well between the electrodes to temporarily trap a DNA molecule. Optimizing the surface chemistry and electrochemical behavior of the device is a necessary (but by no means sufficient) step toward the development of a functional device. In particular, effects to be eliminated are (i) electrochemically induced surface alteration through corrosion or reduction of the electrode surface and (ii) formation of hydrogen or oxygen bubbles inside the pore through water decomposition. Even though our motivation is to solve problems encountered in DNA transistor technology, in this paper we report on generic surface chemistry results. We investigated a variety of electrode-electrolyte-solvent systems with respect to their capability of suppressing water decomposition and maintaining surface integrity. We employed cyclic voltammetry and long-term amperometry as electrochemical test schemes, X-ray photoelectron spectroscopy, atomic force microscopy, and scanning, as well as transmission electron microscopy as analytical tools. Characterized electrode materials include thin films of Ru, Pt, nonstoichiometric TiN, and nonstoichiometric TiN carrying a custom-developed titanium oxide layer, as well as custom-oxidized nonstoichiometric TiN coated with a monolayer of hexadecylphosphonic acid (HDPA). We used distilled water as well as aqueous solutions of poly(ethylene glycol) (PEG-300) and glycerol as solvents. One millimolar KCl was employed as electrolyte in all solutions. Our results show that the HDPA-coated custom-developed titanium oxide layer effectively passivates the underlying TiN layer, eliminating any surface alterations through corrosion or reduction within a voltage window from -2 V to +2 V. Furthermore, we demonstrated that, by coating the custom-oxidized TiN samples with HDPA and increasing the concentration of PEG-300 or glycerol in aqueous 1 mM KCl solutions, water decomposition was suppressed within the same voltage window. Water dissociation was not detected when combining custom-oxidized HDPA-coated TiN electrodes with an aqueous 1 mM KCl-glycerol solution at a glycerol concentration of at least 90%. These results are applicable to any system that requires nanoelectrodes placed in aqueous solution at voltages that can activate electrochemical processes.

Simple and versatile methods to integrate directed self-assembly with optical lithography using a polarity-switched photoresist.

We report novel strategies to integrate block copolymer self-assembly with 193 nm water immersion lithography. These strategies employ commercially available positive tone chemically amplified photoresists to spatially encode directing information into precise topographical or chemical prepatterns for the directed self-assembly of block copolymers. Each of these methods exploits the advantageous solubility and thermal properties of polarity-switched positive tone photoresist materials. Precisely registered, sublithographic self-assembled structures are fabricated using these versatile integration schemes which are fully compatible with current optical lithography patterning materials, processes, and tooling.