Double-Step Machine Learning Based Procedure for HFOs Detection and Classification

Abstract

The need for automatic detection and classification of high-frequency oscillations (HFOs) as biomarkers of the epileptogenic tissue is strongly felt in the clinical field. In this context, the employment of artificial intelligence methods could be the missing piece to achieve this goal. This work proposed a double-step procedure based on machine learning algorithms and tested it on an intracranial electroencephalogram (iEEG) dataset available online. The first step aimed to define the optimal length for signal segmentation, allowing for an optimal discrimination of segments with HFO relative to those without. In this case, binary classifiers have been tested on a set of energy features. The second step aimed to classify these segments into ripples, fast ripples and fast ripples occurring during ripples. Results suggest that LDA applied to 10 ms segmentation could provide the highest sensitivity (0.874) and 0.776 specificity for the discrimination of HFOs from no-HFO segments. Regarding the three-class classification, non-linear methods provided the highest values (around 90%) in terms of specificity and sensitivity, significantly different to the other three employed algorithms. Therefore, this machine-learning-based procedure could help clinicians to automatically reduce the quantity of irrelevant data.

[Retracted] Automatic Detection of High‐Frequency Oscillations Based on an End‐to‐End Bi‐Branch Neural Network and Clinical Cross‐Validation

“…In more recent studies, the automatic detector using the combination of short-time energy and CNN reported in 2019 achieved a sensitivity of 91.26%, a specificity of 91.52%, and a precision of 88.67% in 2700 HFO events from 5 patients [ 21 ]. Sciaraffa et al [ 27 ] proposed a new double-step machine learning method in 2020 to detect HFOs, which achieved a sensitivity of 87.40% and a specificity of 77.60%. Guo et al [ 45 ] developed a hypergraph-based detector to automatically detect HFOs in 2021, which achieved an accuracy of 90.7%, a sensitivity of 80.9%, and a specificity of 96.9%.…”

“…In early studies, researchers mostly focused on the performance of signal detection, so in terms of dividing experimental data, most of them randomly divide training set and test set from the candidate pool [ 21 , 27 , 43 , 45 , 46 ]. In clinical application, when considering a new patient, it is desirable to transfer the a priori knowledge learned from previous existing cases to the judgment of the new patient.…”

HFO Detection in Epilepsy: A Stacked Denoising Autoencoder and Sample Weight Adjusting Factors-Based Method

“…Class imbalance is another issue in the feature classification step of HFO detection, which is caused by the fact that the number of HFOs is significantly less than that of fHFOs in iEEG signals [19], [31]. This class imbalance will sharply drop the classification performance since the class-imbalanced data distribution between HFOs and fHFOs biases classifiers toward the majority class (i.e., the fHFO class) [32].…”

Detection of high-frequency oscillations in electroencephalography: A scoping review and an adaptable open-source framework

“…These detectors often employed multi-stage algorithms, which involved the identification of putative HFO events, extraction of features from these events, and subsequent use of the extracted features to cluster oscillatory events or train the algorithm. Akin to detectors employing time-frequency analysis, variable methods were used for the initial detection of putative HFO events including energy thresholding [54,68] and multi-feature extraction [56,59]. Methods to reject false positive HFO events included analysis of the duration of the event and amplitude threshold.…”