Deep learning for automated epileptiform discharge detection from scalp EEG: A systematic review.

Automated interictal epileptiform discharge (IED) detection has been widely studied, with machine learning methods at the forefront in recent years. As computational resources become more accessible, researchers have applied deep learning (DL) to IED detection with promising results. This systematic review aims to provide an overview of the current DL approaches to automated IED detection from scalp electroencephalography (EEG) and establish recommendations for the clinical research community. We conduct a systematic review according to the PRISMA guidelines. We searched for studies published between 2012 and 2022 implementing DL for automating IED detection from scalp EEG in major medical and engineering databases. We highlight trends and formulate recommendations for the research community by analyzing various aspects: data properties, preprocessing methods, DL architectures, evaluation metrics and results, and reproducibility. The search yielded 66 studies, and 23 met our inclusion criteria. There were two main DL networks, convolutional neural networks in 14 studies and long short-term memory networks in three studies. A hybrid approach combining a hidden Markov model with an autoencoder was employed in one study. Graph convolutional network was seen in one study, which considered a montage as a graph. All DL models involved supervised learning. The median number of layers was 9 (IQR: 5-21). The median number of IEDs was 11 631 (IQR: 2663-16 402). Only six studies acquired data from multiple clinical centers. AUC was the most reported metric (median: 0.94; IQR: 0.94-0.96). The application of DL to IED detection is still limited and lacks standardization in data collection, multi-center testing, and reporting of clinically relevant metrics (i.e. F1, AUCPR, and false-positive/minute). However, the performance is promising, suggesting that DL might be a helpful approach. Further testing on multiple datasets from different clinical centers is required to confirm the generalizability of these methods.

Incidence of stroke subtypes in the North East Melbourne Stroke Incidence Study (NEMESIS): differences between men and women.

BACKGROUND AND PURPOSE: Incidence rates of stroke subtypes may be imprecise when samples are small. We aimed to determine the incidence of stroke subtypes in a large geographically defined population. METHODS: Multiple overlapping sources were used to ascertain all strokes occurring in 22 postcodes (population of 306,631) of Melbourne, Australia, between 1997 and 1999. Stroke subtypes were defined by CT, MRI and autopsy. The Mantel-Haenszel age-adjusted rate ratio (MH RR) was used to compare incidence rates between men and women. RESULTS: We identified 1,421 strokes among 1,337 residents, 1,035 (72.8%) being first-ever strokes. Incidence (number/100,000 population/year), adjusted to the European population 45-84 years, was 197 (95% confidence interval, CI, 169-224) for ischemic stroke (IS), 47 (95% CI 33-60) for intracerebral haemorrhage (ICH) and 19 (95% CI 10-27) for subarachnoid haemorrhage (SAH). Compared with women, men in this age group had a greater incidence of IS (MH RR 1.65, 95% CI 1.39-1.96, p < 0.0001) and ICH (MH RR 1.46, 95% CI 1.01-2.10, p = 0.0420), but lesser rates of SAH (MH RR 0.34, 95% CI 0.16-0.69, p = 0.0031). CONCLUSIONS: In this population-based study, the incidence of IS and ICH was greater among men than women, while women had a greater incidence of SAH. More effort may need to be directed at modifying risk factors for IS and ICH in men.

Greater incidence of both fatal and nonfatal strokes in disadvantaged areas: the Northeast Melbourne Stroke Incidence Study.

BACKGROUND AND PURPOSE: Greater stroke mortality has been reported among lower socioeconomic groups. We aimed to determine whether fatal, nonfatal, and overall stroke incidence varied by socioeconomic status. METHODS: All suspected strokes occurring in 22 postcodes (population of 306,631) of Melbourne, Australia, during a 24-month period between 1997 and 1999 were found and assessed. Multiple overlapping sources were used to ascertain cases with standard clinical definitions for stroke. Socioeconomic disadvantage was assigned in 4 bands from least to greatest using an area-based measure developed by the Australian Bureau of Statistics. RESULTS: Overall stroke incidence (number per 100,000 population per year), adjusted to the European population 45 to 84 years of age, increased with increasing socioeconomic disadvantage: 200 (95% CI, 173 to 228); 251 (95% CI, 220 to 282); 309 (95% CI, 274 to 343); and 366 (95% CI, 329 to 403; chi2 for ranks; P<0.0001). Similar incidence patterns were observed for both fatal and nonfatal stroke. Nonfatal stroke contributed most to this incidence pattern: 146 (95% CI, 122 to 169); 181 (95% CI, 155 to 207); 223 (95% CI, 194 to 252); and 280 (95% CI, 247 to 313; chi2 for ranks; P<0.0001). CONCLUSIONS: In this population-based study, both fatal and nonfatal stroke incidence increased with increasing socioeconomic disadvantage. The greater contributor to this incidence pattern was nonfatal stroke incidence. This may have implications for service provision to those least able to afford it. Area-based identification of those most disadvantaged may provide a simple and effective way of targeting regions for stroke prevention strategies.

Determinants of handicap after stroke: the North East Melbourne Stroke Incidence Study (NEMESIS).

BACKGROUND AND PURPOSE: Handicap, although more relevant to the patient than impairment or disability, has received little attention in people with stroke. The aim of this study was to identify, in an unselected population, factors determining handicap at 2 years after stroke. METHODS: All first-ever cases of stroke in a population of 306 631 over a 1-year period were assessed. Stroke severity, comorbidity, and demographic information was recorded. Among survivors, 2-year poststroke handicap was assessed with the London Handicap Scale. Disability, physical impairment, depression, anxiety, living arrangements, and recurrent stroke at 2 years were documented. If necessary, proxy assessments were obtained, except for mood. Linear regression analyses were performed to identify factors independently associated with handicap. First, all assessments (proxy and nonproxy) were examined; then, the nonproxy assessments were used to examine the effects of mood. RESULTS: Of 266 patients with incident stroke who were alive at 2 years, 226 (85%) were assessed. Significant determinants of handicap on univariable analysis were age, female sex, socioeconomic status, alcohol intake, stroke subtype, initial stroke severity; 2-year physical impairment, disability, depression and anxiety scores; institutionalization; and recurrent stroke. On multivariable analysis, the independent determinants of handicap were age and 2-year physical impairment and disability. In analysis restricted to nonproxy data, depression and anxiety were also independently associated with handicap. CONCLUSIONS: Age, concurrent disability, and physical impairment were more important determinants of handicap than other demographic factors or initial stroke severity. Because depression and anxiety were independently associated with handicap, their treatment may potentially reduce handicap in stroke patients.

Quality of life after stroke: the North East Melbourne Stroke Incidence Study (NEMESIS).

BACKGROUND AND PURPOSE: Health-related quality of life (HRQoL) data are scarce from unselected populations. The aims were to assess HRQoL at 2 years poststroke, to identify determinants of HRQoL in stroke survivors, and to identify predictors at stroke onset of subsequent HRQoL. METHODS: All first-ever cases of stroke in a population of 306 631 over a 1-year period were assessed. Stroke severity, comorbidity, and demographic information were recorded. Two-year poststroke HRQoL was assessed using the Assessment of Quality of Life (AQoL) instrument (deceased patients score=0). Handicap, disability, physical impairment, depression, anxiety, living arrangements, and recurrent stroke at 2 years were documented. If necessary, proxy assessments were obtained, except for mood. Linear regression analyses were performed to identify factors independently associated with HRQoL. RESULTS: Of 266 incident cases alive at 2 years, 225 (85%) were assessed. The mean AQoL utility score for all survivors was 0.47 (95% CI, 0.42 to 0.52). Almost 25% of survivors had a score of < or =0.1. The independent determinants of HRQoL in survivors were handicap, physical impairment, anxiety and depression, disability, institutionalization, dementia, and age. The factors present at stroke onset that independently predicted HRQoL at 2 years poststroke were age, female sex, initial NIHSS score, neglect, and low socioeconomic status. CONCLUSIONS: A substantial proportion of stroke survivors have very poor HRQoL. Interventions targeting handicap and mood have the potential to improve HRQoL independently of physical impairment and disability.

Quality of stroke care within a hospital: effects of a mobile stroke service.

OBJECTIVE: An Australian stroke services study (SCOPES) has developed a framework to compare different forms of acute stroke services, the gold standard being localised stroke units. We aimed to use this framework to assess changes in the quality of stroke care over time as a sequential audit process. DESIGN AND SETTING: A retrospective medical record audit comparing 100 sequential stroke admissions (July 2002 to June 2003) two years after institution of a mobile stroke service (MSS) with 100 historical controls (September 1998 to October 1999) at a 260-bed hospital in Melbourne. The MSS results were also compared with stroke units in SCOPES. MAIN OUTCOME MEASURES: Adherence to quality indicators and standard measures of outcome (complications, length of stay and discharge disability) after implementing the MSS. RESULTS: Significant improvements were seen in prophylaxis for deep-vein thrombosis, incontinence management, premorbid function documentation, frequent neurological observations and early occupational therapy. The MSS demonstrated fewer severe complications (9% versus 24%; P = 0.004), reduced median length of stay (discharged patients: 12.0 days versus 18.5 days; P = 0.003) and more patients were independent at discharge (32% versus 9%; P < 0.001). Comparison with SCOPES stroke units showed our MSS could improve in incontinence management and appropriate use of antiplatelet therapy. CONCLUSION: Institution of the MSS was associated with improvements in the quality of stroke care. This study demonstrates application of an audit procedure for quality improvement in hospital stroke management and the potential to improve stroke services in smaller centres.

Stroke units, tissue plasminogen activator, aspirin and neuroprotection: which stroke intervention could provide the greatest community benefit?

BACKGROUND: Although a number of acute stroke interventions are of proven efficacy, there is uncertainty about their community benefits. We aimed to assess this within a defined population. METHODS: Eligibility for tissue plasminogen activator (tPA), aspirin, stroke unit management and neuroprotection were assessed among incident stroke cases within the community-based North East Melbourne Stroke Incidence Study. RESULTS: Among 306,631 people, there were 645 incident strokes managed in hospital. When eligible patients were extrapolated to the Australian population, for every 1,000 cases, 46 (95% CI 17-69) could have been saved from death or dependency with stroke unit management, 6 (95% CI 1-11) by using aspirin, 11 (95% CI 5-17) or 10 (95% CI 3-16) by using tPA at 3 and 6 h, respectively. CONCLUSIONS: Although tPA is the most potent intervention, management in stroke units has the greatest population benefit and should be a priority.