Automated labelling of epileptic seizures on electroencephalograms is an essential interdisciplinary task of diagnostics. Traditional machine learning approaches operate in a supervised fashion requiring complex pre-processing procedures that are usually labour intensive and time-consuming. The biggest issue with the analysis of electroencephalograms is the artefacts caused by head movements, eye blinks, and other non-physiological reasons. Similarly to epileptic seizures, artefacts produce rare high-amplitude spikes on electroencephalograms, complicating their separability. We suggest that artefacts and seizures are rare events; therefore, separating them from the rest data seriously reduces information for further processing. Based on the occasional nature of these events and their distinctive pattern, we propose using anomaly detection algorithms for their detection. These algorithms are unsupervised and require minimal pre-processing. In this work, we test the possibility of an anomaly (or outlier) detection algorithm to detect seizures. We compared the state-of-the-art outlier detection algorithms and showed how their performance varied depending on input data. Our results evidence that outlier detection methods can detect all seizures reaching 100% recall, while their precision barely exceeds 30%. However, the small number of seizures means that the algorithm outputs a set of few events that could be quickly classified by an expert. Thus, we believe that outlier detection algorithms could be used for the rapid analysis of electroencephalograms to save the time and effort of experts.

(1) In this work, a set of deep autoencoders with different numbers of layers and layer sizes were trained using a set of open access hand tracking datasets. Examining the results allowed to determine optimal number of layers for deep autoencoders and augmentation parameters for denoising autoencoders. Additionally, variations of autoencoders reconstruction loss depending on hidden unit size and latent space size were observed. (2) Results. Trained autoencoders showed better results in representation extraction than PCA method. A positive relationship between autoencoder depth and model performance was also proven. Future work will be aimed at applying obtained results to transfer learning problems.

The aim of the study was to identify potential predictors of functional outcome (FO) in patients with subtypes of ischemic stroke (IS) who did not receive reperfusion therapy. Materials and methods. A prospective study included 229 patients with ischemic stroke divided . into three groups based on the IS subtype: Group 1 — 84 patients with cardioembolic IS; Group 2 — 65 patients with atherothrombotic IS; Group 3 — 80 patients with lacunar IS. Changes in the modified Rankin Scale (mRS) scores were considered as FO criteria calculated as the difference between the scores on admission and on the 21 ^st day after IS onset — ∆mRS. In order to optimize the performance of the machine learning (ML) model, a binary FO approach was chosen for assessment on the 21 ^st day after IS onset: mRS ≥ 3 scores corresponded to an unfavorable non-lethal outcome, and mRS = 0–2 scores corresponded to a favorable FO. We analyzed the interrelation with FO (correlation coefficient, r) and the predictive ability (ML (decision tree), information gain, i. g.) of 29 parameters, including demographic features; comorbidities; instrumental examination findings; NIHSS, BI, CDR scores; serum concentrations of cytokines on the 2 ^nd day of hospital stay. Results. The following significant ( P <0.0001) predictors of unfavorable non-lethal FO were identified: female sex (i. g. = 0.346), recurrent IS (i. g = 0.248), diabetes mellitus (i. g. = 0.442), and CXCL2 concentration (i.g. = 0.306) in Group 1; WMHs severity (i. g. = 0.206), diabetes mellitus (i. g. = 0.340), content of CCL2 (i. g. = 0.116), CCL3 (i. g. = 0.202) and CCL23 (i. g. = 0.101) in Group 2; age (i. g. = 0.106), 2 ^nd –3 ^rd degree obesity (i. g. = 0.150), WMHs severity (i. g. = 0.300), CXCL5 content (i. g. = 0.143) and MIF (i. g. = 0.145) in Group 3. Concentrations of CCL25 (i. g. = 0.108) and IL-6 (i. g. = 0.401) were found as predictors of favorable FO ( P <0.0001) in Group 1; 1 ^st degree obesity (i. g. = 0.118) and TNF-α concentration (i. g. = 0.211) in Group 2; arterial hypertension (AH) (i. g. = 0.113) and 1 ^st degree obesity in Group 3. Conclusion. Study results made evident the variances in combination of factors affecting FO, depending on IS pathogenetic subtype. Despite undoubtful value of the data obtained, further research is needed to expand the potentiality in predicting acute IS outcome and confirm the relevance of identified markers.