pmcid stringlengths 9 11 | sentence stringlengths 1 4.74k | confidence float64 0.5 1 | is_ignorance bool 2
classes | model_used stringclasses 1
value |
|---|---|---|---|---|
PMC467014 | We would like to thank the authors for their interest in our article titled “Evaluation of ganglion cell complex and retinal nerve fiber layer thinning in epilepsy patients. | 0.9928 | false | biomedbert-dasha |
PMC467014 | ” We wanted to clarify the questions they mentioned. | 0.9909 | false | biomedbert-dasha |
PMC467014 | Vitamin (folate and B12) values of the patients included in the study were checked. | 0.9962 | false | biomedbert-dasha |
PMC467014 | Those who were within normal limits were included in the study. | 0.9976 | false | biomedbert-dasha |
PMC467014 | Neurologic examinations of the participants were performed. | 0.9974 | false | biomedbert-dasha |
PMC467014 | Those with any neurologic disease (including migraine) were excluded from the study. | 0.9982 | false | biomedbert-dasha |
PMC467014 | Smoking history of the participants was not questioned. | 0.9929 | false | biomedbert-dasha |
PMC467014 | This may be considered a limitation of our study. | 0.9965 | false | biomedbert-dasha |
PMC467014 | However, the effect of smoking on the retinal nerve fiber layer (RNFL) and ganglion cell complex (GCC) is controversial. | 0.9955 | true | biomedbert-dasha |
PMC467014 | Although there are studies showing thinning, there are studies showing no change. | 0.9923 | false | biomedbert-dasha |
PMC467014 | Some studies have also shown that smoking affects one but not the other. | 0.9623 | false | biomedbert-dasha |
PMC467014 | It is statistical that GCC is not significant in the inferior 3 mm. | 0.9948 | false | biomedbert-dasha |
PMC467014 | Measurements of the patient group (116.74 ± 13.47) and the control group (121.74 ± 8.90) revealed that the standard deviation was higher in the patient group. | 0.9989 | false | biomedbert-dasha |
PMC467014 | This difference seems to have affected the significance. | 0.9971 | false | biomedbert-dasha |
PMC467014 | As our study was not suitable for drug evaluation in terms of number of patients and methodology, we grouped the patients as monotherapy and polytherapy recipients and compared them. | 0.9983 | false | biomedbert-dasha |
PMC467014 | We wrote in the limitations section that long-term and large-participant studies are needed to evaluate the effects of drugs and epilepsy on the retina. | 0.9813 | true | biomedbert-dasha |
PMC467014 | Financial support and sponsorship: Nil. | 0.994 | false | biomedbert-dasha |
PMC467014 | Conflicts of interest: There are no conflicts of interest. | 0.9964 | false | biomedbert-dasha |
PMC7613524 | The takeover of Afghanistan by the Taliban in August, 2021, and the associated collapse of governmental institutions has led to innumerable security and humanitarian concerns, particularly for women and girls. | 0.9961 | false | biomedbert-dasha |
PMC7613524 | Even before this event, more than 550 000 Afghans had been internally displaced during 2021. | 0.9979 | false | biomedbert-dasha |
PMC7613524 | At the end of 2020, 3 million Afghans were internally displaced, and 2·6 million Afghan people were recognised as refugees or asylum seekers; most of them hosted in neighbouring Pakistan and Iran. | 0.9987 | false | biomedbert-dasha |
PMC7613524 | The hastily arranged mass exodus, as international service personnel withdrew, has further spread Afghan refugees worldwide. | 0.9979 | false | biomedbert-dasha |
PMC7613524 | Although many people who were forced to leave Afghanistan are capable, literate, and determined, adjusting to new societies and lifestyles is not straightforward. | 0.9889 | false | biomedbert-dasha |
PMC7613524 | Forcibly displaced people, within or outside Afghanistan, often have restricted access to security, shelter, and health-care facilities and inadequate nutrition. | 0.9955 | false | biomedbert-dasha |
PMC7613524 | Some individuals are also at high risk of persecution and death. | 0.9983 | false | biomedbert-dasha |
PMC7613524 | Among people living in refugee camps, the most reported neurological condition is epilepsy. | 0.9983 | false | biomedbert-dasha |
PMC7613524 | Despite affecting over 50 million people worldwide and being treatable with affordable antiseizure medications in most people, epilepsy is a deeply stigmatised condition. | 0.9917 | false | biomedbert-dasha |
PMC7613524 | This stigma is partly attributable to long-standing misconceptions about the neurological basis of seizures and other traditional beliefs, for example, that epilepsy is contagious. | 0.9948 | false | biomedbert-dasha |
PMC7613524 | Convulsions are also unpredictable and difficult to witness, leading to fear and anxiety, including in health-care workers helping people who have been displaced. | 0.9979 | false | biomedbert-dasha |
PMC7613524 | Asylum seekers can, therefore, be wary of declaring an epilepsy diagnosis, fearing that this will prejudice refugee status applications. | 0.999 | false | biomedbert-dasha |
PMC7613524 | All of these factors, coupled with, for example, a scarcity of medically trained personnel with appropriate expertise, contribute to forcibly displaced people with epilepsy receiving suboptimal care. | 0.999 | false | biomedbert-dasha |
PMC7613524 | As the turmoil in Afghanistan continues to unfold, health-care systems should adapt to the increasing needs of forcibly displaced populations. | 0.9956 | false | biomedbert-dasha |
PMC7613524 | Health-care workers should be trained in the essential management of seizures to help people in the acute setting. | 0.9984 | false | biomedbert-dasha |
PMC7613524 | Improved knowledge about epilepsy, encouraging improved adherence to antiseizure medications, and destigmatisation will help improve prognosis. | 0.9985 | false | biomedbert-dasha |
PMC7613524 | A transcultural approach seems crucial; for example, Afghan refugees and asylum seekers have previously reported feeling isolated in their contact with host–country institutions that, superficially, seem culturally similar. | 0.9952 | false | biomedbert-dasha |
PMC7613524 | Awareness of common causes of epilepsy in Afghanistan, differential diagnoses for seizures, and the context that people are leaving, with special attention to hardships faced by women and girls, should help forcibly displaced Afghans with epilepsy and people who develop seizures to integrate into their receiving countr... | 0.9972 | false | biomedbert-dasha |
PMC7613524 | We declare no competing interests. | 0.9954 | false | biomedbert-dasha |
PMC7613668 | This work explores the potential utility of neural network classifiers for real-time classification of field-potential based biomarkers in next-generation responsive neuromodulation systems. | 0.9964 | false | biomedbert-dasha |
PMC7613668 | Compared to classical filter-based classifiers, neural networks offer an ease of patient-specific parameter tuning, promising to reduce the burden of programming on clinicians. | 0.9986 | false | biomedbert-dasha |
PMC7613668 | The paper explores a compact, feed-forward neural network architecture of only dozens of units for seizure-state classification in refractory epilepsy. | 0.9972 | false | biomedbert-dasha |
PMC7613668 | The proposed classifier offers comparable accuracy to filterclassifiers on clinician-labeled data, while reducing detection latency. | 0.9972 | false | biomedbert-dasha |
PMC7613668 | As a trade-off to classical methods, the paper focuses on keeping the complexity of the architecture minimal, to accommodate the on-board computational constraints of implantable pulse generator systems. | 0.995 | false | biomedbert-dasha |
PMC7613668 | Clinical relevance—A neural network-based classifier is presented for responsive neurostimulation, with comparable accuracy to classical methods at reduced latency. | 0.9968 | false | biomedbert-dasha |
PMC7613668 | Deep brain stimulation (DBS) first received approval for the symptomatic treatment of Parkinson’s disease in 1997. | 0.9985 | false | biomedbert-dasha |
PMC7613668 | While similar in design to cardiac pacemakers, the implantable pulse generators (IPG) of the time offered only an open-loop form of therapy, with typically a single stimulation pattern, set by a clinician for each patient. | 0.9984 | false | biomedbert-dasha |
PMC7613668 | Real-time seizure detection and responsive neurostimulation (RNS) was first attempted using a computer-in-the-loop system by Gotman et al. in 1976, it was Osorio et al. in 1998,, who introduced the more widely studied filter-based spectral biomarker detectors to the field of epilepsy research. | 0.9976 | false | biomedbert-dasha |
PMC7613668 | With the continued development of IPGs and the maturation of low-power microprocessor technology, the first RNS system for epilepsy received approval for pre-clinical use in 2014. | 0.9979 | false | biomedbert-dasha |
PMC7613668 | This system from Neuropace had the capabilities to sense bioelectric signals, and choose stimulation programs based on clinician-configured classification state. | 0.9977 | false | biomedbert-dasha |
PMC7613668 | Filter-based spectral-feature detectors have since been used successfully in other conditions, most notably for tremor suppression in Parkinson’s disease, through the discovery of beta oscillations as a correlate of disease state. | 0.999 | false | biomedbert-dasha |
PMC7613668 | However, the smaller (1 μVrms) signal size of beta oscillations, compared to epileptiform activity (10 μVrms), made deploying the detector algorithm in IPGs challenging due to the presence of stimulation and other artifacts. | 0.9965 | false | biomedbert-dasha |
PMC7613668 | Contemporary work focuses on improving the robustness of the signal chains to enable simultaneous sensing and stimulation, thus true closed-loop operation across targeted diseases,. | 0.9735 | false | biomedbert-dasha |
PMC7613668 | Examples include the Medtronic Percept and the Picostim-DyNeuMo research systems,. | 0.9968 | false | biomedbert-dasha |
PMC7613668 | A complementary avenue of refinement is the use of feedforward predictors for adapting stimulation based on periodicities of disease state and patient needs, such the circadian scheduling of the SenTiva system from LivaNova or the Picostim-DyNeuMo. | 0.9889 | false | biomedbert-dasha |
PMC7613668 | Taking advantage of more long term, weekly or even monthly rhythms are being investigated for epilepsy management. | 0.995 | false | biomedbert-dasha |
PMC7613668 | Patient-specific filter design, while possible to aid with software, can be a complex problem, likely to limit both clinician-capacity and patient-throughput. | 0.9947 | false | biomedbert-dasha |
PMC7613668 | Establishing and validating a neural-network (NN) training pipeline based on clinician-labeled data could offer a systematic classifier tuning process. | 0.9976 | false | biomedbert-dasha |
PMC7613668 | Networks could be pre-trained on aggregate data from multiple patients, and refined based on individual labeled data at the point of deployment. | 0.9989 | false | biomedbert-dasha |
PMC7613668 | Of course, due to the black-box nature of neural network classifiers, extensive validation work will be required to establish safety before first-in-human studies. | 0.9866 | true | biomedbert-dasha |
PMC7613668 | Advances in interpretable deep learning could facilitate building trust in NN–classifiers for medical use. | 0.9922 | false | biomedbert-dasha |
PMC7613668 | Liu et al. demonstrated the feasibility of deploying high accuracy classifiers for seizure detection on modern microprocessors (ARM® Cortex-M4), through model compression and quantization techniques, showcasing several advanced NN topologies. | 0.9977 | false | biomedbert-dasha |
PMC7613668 | This paper is meant as an initial study to bring focus to the fundamental challenge of NN classifiers: computational cost. | 0.9441 | false | biomedbert-dasha |
PMC7613668 | As state-of-the-art deep neural networks reach ever increasing model sizes,, we aim to explore whether lean NNs of only dozens of units could in fact compete in accuracy with classical, filter-based systems for bioelectric signal classification. | 0.9916 | false | biomedbert-dasha |
PMC7613668 | To establish a baseline for performance as well as for computational cost, we used a classical band-power estimation filter chain to detect epileptiform discharges,, that we have previously deployed with success in the Picostim-DyNeuMo experimental IPG system. | 0.9984 | false | biomedbert-dasha |
PMC7613668 | The processing steps of this method are shown in the top panel of Fig. 1. | 0.9921 | false | biomedbert-dasha |
PMC7613668 | While this algorithm is computationally efficient and has a very favorable memory footprint (refer to Table I), the demodulated envelope signal, thus the detector output, will always lag the input signal to reduce output ripple – irrespective of processing speed. | 0.999 | false | biomedbert-dasha |
PMC7613668 | This trade-off arises from the very nature of causal filtering, and is necessary to prevent rapid switching of the detector output for input signals near the classification threshold. | 0.9989 | false | biomedbert-dasha |
PMC7613668 | The reference classifier was configured as follows. | 0.9969 | false | biomedbert-dasha |
PMC7613668 | Our band-pass stage was an 8 – 22 Hz, 4th order Butterworth filter, with a Direct Form I IIR implementation (16-bit coefficients, 32-bit accumulators). | 0.9968 | false | biomedbert-dasha |
PMC7613668 | Envelope demodulation was achieved using an exponential moving average filter with a decay-coefficient of 32 samples. | 0.9978 | false | biomedbert-dasha |
PMC7613668 | The filter chain, and all other classifiers were designed to operate at a sampling rate of 256 Hz. | 0.9977 | false | biomedbert-dasha |
PMC7613668 | In our search for a low-complexity classifier for time series input, we explored two main NN families. | 0.9982 | false | biomedbert-dasha |
PMC7613668 | (1) Multi-Layer Perceptrons (MLP) are the simplest, and oldest family or of artificial neural networks, where the input vector is connected to ‘hidden’ layers of feedforward units, condensing information into an output unit. | 0.9982 | false | biomedbert-dasha |
PMC7613668 | This architecture is shown in the bottom panel of Fig. 1. | 0.9908 | false | biomedbert-dasha |
PMC7613668 | (2) As a step up in complexity, Convolutional Neural Network (CNN) introduce a convolution layer, also known as a filter bank, between the input vector and the neural layers as an initial extra step. | 0.9984 | false | biomedbert-dasha |
PMC7613668 | The input to our networks is formed by a windowed set of past time samples of the local field potential (LFP) signal. | 0.998 | false | biomedbert-dasha |
PMC7613668 | The output signal, calculated once for each complete window of samples, is thresholded into a binary label. | 0.9958 | false | biomedbert-dasha |
PMC7613668 | We denote this classifier the ‘standalone MLP’ model. | 0.9988 | false | biomedbert-dasha |
PMC7613668 | Recurrent neural networks, an otherwise natural choice for processing time series data, were dismissed from consideration as recurrence necessitates the introduction of dynamic state variables, which significantly increases memory footprint. | 0.9985 | false | biomedbert-dasha |
PMC7613668 | Without recurrence, we introduced coherence into our classifier in a different way. | 0.9988 | false | biomedbert-dasha |
PMC7613668 | We settled on requiring a consensus of three subsequent outputs from the NN to define our final output label, providing the ‘adjusted MLP’ model. | 0.9991 | false | biomedbert-dasha |
PMC7613668 | Our raw dataset consisted of LFP signals recorded from two patients, for a combined 24 hours, with 30 professionally labeled events of clinical significance. | 0.9982 | false | biomedbert-dasha |
PMC7613668 | The recordings were resampled to a 256 Hz sampling frequency for uniformity. | 0.9972 | false | biomedbert-dasha |
PMC7613668 | As seizures are comparatively rare events scattered among very long periods of normal activity, we decided to introduce class imbalance into our training sets to best prepare the NNs for real-life use. | 0.9989 | false | biomedbert-dasha |
PMC7613668 | The training set was biased towards negative samples in a 3:1 ratio, based on clinician annotations. | 0.9994 | false | biomedbert-dasha |
PMC7613668 | The dataset was split in the common 70:30 ratio between a training and a validation set. | 0.9989 | false | biomedbert-dasha |
PMC7613668 | Network weights and biases were quantized to 8-bit integers. | 0.9971 | false | biomedbert-dasha |
PMC7613668 | Neural networks were modeled and trained in Tensorflow Lite version 2.7.0, using an Intel Core i7 CPU with 16 GB of RAM. | 0.9974 | false | biomedbert-dasha |
PMC7613668 | Embedded performance was tested on an Arduino Nano 33 BLE Sense evaluation board for the nRF52840 ARM Cortex-M4F microprocessor. | 0.995 | false | biomedbert-dasha |
PMC7613668 | Fig. 2 shows the performance of our CNN and MLP classifiers. | 0.9972 | false | biomedbert-dasha |
PMC7613668 | While the CNN outperforms the standalone MLP, it performs with similar accuracy to the adjusted MLP model above 80% true positive rate (TPR). | 0.998 | false | biomedbert-dasha |
PMC7613668 | For safe use, the operating point of a seizure detection system should be biased towards high TPR – missed seizures (false negatives) pose significantly more risk to the patient than false positives, which merely result in unnecessary stimulation. | 0.9992 | false | biomedbert-dasha |
PMC7613668 | Overall, in targeting resource constrained IPGs, we judged the minor edge of the CNN insufficient to justify the added computational burden of the convolutional layer. | 0.9988 | false | biomedbert-dasha |
PMC7613668 | The performance of the MLP model, when trained on a given dataset, is primarily determined by two hyperparameters: the number of timepoints in the input window, and the hidden layer’s size. | 0.9976 | false | biomedbert-dasha |
PMC7613668 | We found that varying the number of hidden layers had very modest effects on accuracy (not shown in this paper). | 0.9987 | false | biomedbert-dasha |
PMC7613668 | Fig. 3 systematically explores the effect of the two key hyperparameters on the classification error of a single output, single hidden layer MLP model. | 0.9987 | false | biomedbert-dasha |
PMC7613668 | As expected, the network requires a certain size and complexity to encode a feature space sufficient for reliable classification, though increasing the number of units in either layer beyond a certain point leads to diminishing returns. | 0.9987 | false | biomedbert-dasha |
PMC7613668 | To select one of the possible models from the error surface, one could define a scoring scheme including network size, computational time and the loss itself, to make an educated choice, however, this is beyond the scope of this paper. | 0.9816 | false | biomedbert-dasha |
PMC7613668 | Favoring low complexity, we settled on using a 20-point input window and 8 hidden neurons, in the ‘transition zone’ of the error surface. | 0.9986 | false | biomedbert-dasha |
PMC7613668 | The next step is to compare our best MLP result to the baseline classifier. | 0.989 | false | biomedbert-dasha |
PMC7613668 | The ROC of Fig. 2 reveals that a well-tuned filter chain outperforms the small MLP model below 60% false positive rate, beyond which they converge in accuracy. | 0.9983 | false | biomedbert-dasha |
End of preview. Expand in Data Studio
README.md exists but content is empty.
- Downloads last month
- -