Dataset Information

A High Accuracy Electrographic Seizure Classifier Trained Using Semi-Supervised Labeling Applied to a Large Spectrogram Dataset.

ABSTRACT: The objective of this study was to explore using ECoG spectrogram images for training reliable cross-patient electrographic seizure classifiers, and to characterize the classifiers' test accuracy as a function of amount of training data. ECoG channels in ∼138,000 time-series ECoG records from 113 patients were converted to RGB spectrogram images. Using an unsupervised spectrogram image clustering technique, manual labeling of 138,000 ECoG records (each with up to 4 ECoG channels) was completed in 320 h, which is an estimated 5 times faster than manual labeling without ECoG clustering. For training supervised classifier models, five random folds of data were created; with each fold containing 72, 18, and 23 patients' data for model training, validation and testing respectively. Five convolutional neural network (CNN) architectures, including two with residual connections, were trained. Cross-patient classification accuracies and F₁ scores improved with model complexity, with the shallowest 6-layer model (with ∼1.5 million trainable parameters) producing a class-balanced seizure/non-seizure classification accuracy of 87.9% on ECoG channels and the deepest ResNet50-based model (with ∼23.5 million trainable parameters) producing a classification accuracy of 95.7%. The trained ResNet50-based model additionally had 93.5% agreement in scores with an independent expert labeller. Visual inspection of gradient-based saliency maps confirmed that the models' classifications were based on relevant portions of the spectrogram images. Further, by repeating training experiments with data from varying number of patients, it was found that ECoG spectrogram images from just 10 patients were sufficient to train ResNet50-based models with 88% cross-patient accuracy, while at least 30 patients' data was required to produce cross-patient classification accuracies of >90%.

SUBMITTER: Barry W

PROVIDER: S-EPMC8273175 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Background Radiotracer extravasations, caused largely by faulty tracer injections, can occur in up to 23% of 18F-fluorodeoxyglucose (FDG) PET/CT scans and negatively impact radiological review and tracer quantification. Conventional radiological assessment of extravasation severity on PET has limited performance (e.g., extravasations frequently resolve before scanning) and practical drawbacks. In this study, we develop a new topical detector-based FDG extravasation severity classifier, calibrated from semi-quantitative PET measurements, and assess its performance on human subjects. Methods A retrospective study examined patients whose FDG injections had been monitored as part of their standard workup for PET/CT imaging. Topical uncollimated gamma ray detectors were applied proximal to the injection site and on the same location on the opposing arm, and readings were acquired continuously during radiotracer uptake. Patients were imaged with their arms in the PET field of view and total extravasation activity quantified from static PET images through a volume of interest approach. The image-derived activities were considered ground truth and used to calibrate and assess quantification of topical detector readings extrapolated to the start of PET imaging. The classifier utilizes the calibrated detector readings to produce four extravasation severity classes: none, minor, moderate, and severe. In a blinded study, a radiologist qualitatively labeled PET images for extravasation severity using the same classifications. The radiologist’s interpretations and topical detector classifications were compared to the ground truth PET results. Results Linear regression of log-transformed image-derived versus topical detector tracer extravasation activity estimates showed a strong correlation (R2 = 0.75). A total of 24 subject scans were cross-validated with the quantitatively based classifier through a leave-one-out methodology. For binary classification (none vs. extravasated), the topical detector classifier had the highest overall diagnostic performance for identifying extravasations. Specificity, sensitivity, accuracy, and positive predictive value were 100.0%, 80.0%, 95.8%, and 100.0%, respectively, for the topical detector classifier and 31.6%, 100.0%, 45.8%, and 27.8%, respectively, for the radiological analysis. The topical detector classifier, with an optimal detection threshold, produced a significantly higher Matthews correlation coefficient (MCC) than the radiological analysis (0.87 vs. 0.30). Conclusions The topical detector binary classifier, calibrated using quantitative static PET measurements, significantly improves extravasation detection compared to qualitative image analysis. Supplementary Information The online version contains supplementary material available at 10.1186/s40658-022-00488-6.

Project description:ObjectiveRecording seizures using personal seizure diaries can be challenging during everyday life and many seizures are missed or mis-reported. People living with epilepsy could benefit by having a more accurate and objective wearable EEG system for counting seizures that can be used outside of the hospital. The objective of this study was to (1) determine which seizure types can be electrographically recorded from the scalp below the hairline, (2) determine epileptologists' ability to identify electrographic seizures from single-channels extracted from full-montage wired-EEG, and (3) determine epileptologists' ability to identify electrographic seizures from Epilog, a wireless single-channel EEG sensor.MethodsEpilog sensors were worn concurrently during epilepsy monitoring unit (EMU) monitoring. During standard-of-care review, epileptologists were asked if the electrographic portion of the seizure was visible on single channels of wired electrodes at locations proximal to Epilog sensors, and if focal-onset, which electrode was closest to the focus. From these locations, single channels of EEG extracted from wired full-montage EEG and the proximal Epilog sensor were presented to 3 blinded epileptologists along with markers for when known seizures occurred (taken from the standard-of-care review). Control segments at inter-ictal times were included as control. The epileptologists were asked whether a seizure event was visible in the single channel EEG record at or near the marker.ResultsA total of 75 seizures were recorded from 22 of 40 adults that wore Epilog during their visit to the EMU. Epileptologists were able to visualize known seizure activity on at least one of the wired electrodes proximal to Epilog sensors for all seizure events. Epileptologists accurately identified seizures in 71% of Epilog recordings and 84% of single-channel wired recordings and were 92% accurate identifying seizures with Epilog when those seizures ended in a clinical convulsion compared to those that did not (>55%).ConclusionsEpileptologists are able to visualize seizure activity on single-channels of EEG at locations where Epilog sensors are easily placed on the scalp below hairline. Manual review of seizure annotations can be done quickly and accurately (>70% TP and >98% PPV) on single-channel EEG data. Reviewing single-channel EEG is more accurate than what has been reported in the literature on self-reporting seizures in seizure diaries, the current standard of care for seizure counting outside of the EMU.SignificanceWearable EEG will be important for seizure monitoring outside of the hospital. Epileptologists can accurately identify seizures in single-channel EEG, better than patient self-reporting in diaries based on the literature. Automated or semi-automated seizure detection on single channels of EEG could be used in the future to objectively count seizures to complement the standard of care outside of the EMU without the overt burden upon epileptologist review.

Project description:A lot of efforts have been made to understand the structure and function of neocortical circuits. In fact, a promising way to understand the functions of cortical circuits is the classification of the neural types, based on their different properties. Recent studies focused on applying modern computational methods to classify neurons based on molecular, morphological, physiological, or mixed of these criteria. Although there are studies in the literature on in vitro/vivo extracellular or in vitro intracellular recordings, a study on the classification of neuronal types using in vivo whole-cell patch-clamp recordings is still lacking. We thus proposed a novel semi-supervised classification method based on waveform shape of neurons' spikes using in vivo whole-cell patch-clamp recordings. We, first, detected spike candidates. Then discriminative features were extracted from the time samples of the spikes using discrete cosine transform. We then extracted the center of clusters using fuzzy c-mean clustering and finally, the neurons were classified using the minimum distance classifier. We distinguished three types of neurons: excitatory pyramidal cells (Pyr) and two types of inhibitory neurons: GABAergic- parvalbumin positive (PV), and somatostatin positive (SST) non-pyramidal cells in layer II/III of the mice primary visual cortex. We used 10-fold cross validation in our study. The classification accuracy for PV, Pyr, and SST was 91.59 ± 1.69, 97.47 ± 0.67, and 89.06 ± 1.99, respectively. Overall, the algorithm correctly classified 92.67 ± 0.54% of the cells, confirming the relative robustness of the discriminant functions. The performance of the method was further assessed on in vitro recordings by using a pool of 50 neurons from Allen institute Cell Types Database (5 major subtypes of neurons: Pyr, PV, SST, 5HT3a, and vasoactive intestinal peptide (VIP) cells). Its overall accuracy was 84.13 ± 0.81% on this data set using cross validation framework. The proposed algorithm is thus a promising new tool in recognizing cell's type with high accuracy in laboratories using in vivo/vitro whole-cell patch-clamp recording technique. The developed programs and the entire dataset are available online to interested readers.

Dataset Information

A High Accuracy Electrographic Seizure Classifier Trained Using Semi-Supervised Labeling Applied to a Large Spectrogram Dataset.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets