Detection of cortical EEG arousals from a Neural Network (NN) analysis
Traditionally EEG sleep fragmentation is scored according to simple arbitrary thresholds which ignore large amounts of EEG information Computer based analyses may improve this situation. To begin assessing this approach we used a neural network system that provides a second by second output of sleep...
| Main Authors: | , , , |
|---|---|
| Format: | Journal article |
| Language: | English |
| Published: |
1996
|
| _version_ | 1797091435027103744 |
|---|---|
| author | Bennett, L Stradling, JR Barbour, C Davies, R |
| author_facet | Bennett, L Stradling, JR Barbour, C Davies, R |
| author_sort | Bennett, L |
| collection | OXFORD |
| description | Traditionally EEG sleep fragmentation is scored according to simple arbitrary thresholds which ignore large amounts of EEG information Computer based analyses may improve this situation. To begin assessing this approach we used a neural network system that provides a second by second output of sleep 'depth' (Questar, Oxford Medical, UK) to attempt to identify ASDA EEG arousals. 8 men with severe obstructive sleep apnoea were studied. EEG from 20 minutes of continuous OSA with obvious EEG arousal was analysed with the neural network system and then computer processed for arousal detection. By trial and error, we devised an algorithm looking for changes above a baseline. The baseline was defined as the second highest value of a 10 second moving rank filter against which we compared the next 5 one-second data points of EEG. Events were scored 5 times depending on whether 1, 2, 3, 4, or 5 of these points exceeded the baseline (X axis). We compared neural network events with ASDA arousals identifying events as true positive, false positive or false negative. ASDA EEG arousals were consensus scored by 2 scorers. (Graph Presented) When 1 prospective observation exceeded the baseline our algorithm detected 229 of 237 (97%) ASDA arousals with 68 false positives. When 3 prospective points exceeded the filter threshold it detected 198 of 237(84%) ASDA arousals with 19 false positives. Neural network EEG analysis can be post processed to detect some ASDA arousals. It also detects other events which may represent arousal from sleep not detected using the ASDA criteria. |
| first_indexed | 2024-03-07T03:33:01Z |
| format | Journal article |
| id | oxford-uuid:bb5b134d-41c6-4397-afb0-bccbc195caea |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T03:33:01Z |
| publishDate | 1996 |
| record_format | dspace |
| spelling | oxford-uuid:bb5b134d-41c6-4397-afb0-bccbc195caea2022-03-27T05:16:19ZDetection of cortical EEG arousals from a Neural Network (NN) analysisJournal articlehttp://purl.org/coar/resource_type/c_dcae04bcuuid:bb5b134d-41c6-4397-afb0-bccbc195caeaEnglishSymplectic Elements at Oxford1996Bennett, LStradling, JRBarbour, CDavies, RTraditionally EEG sleep fragmentation is scored according to simple arbitrary thresholds which ignore large amounts of EEG information Computer based analyses may improve this situation. To begin assessing this approach we used a neural network system that provides a second by second output of sleep 'depth' (Questar, Oxford Medical, UK) to attempt to identify ASDA EEG arousals. 8 men with severe obstructive sleep apnoea were studied. EEG from 20 minutes of continuous OSA with obvious EEG arousal was analysed with the neural network system and then computer processed for arousal detection. By trial and error, we devised an algorithm looking for changes above a baseline. The baseline was defined as the second highest value of a 10 second moving rank filter against which we compared the next 5 one-second data points of EEG. Events were scored 5 times depending on whether 1, 2, 3, 4, or 5 of these points exceeded the baseline (X axis). We compared neural network events with ASDA arousals identifying events as true positive, false positive or false negative. ASDA EEG arousals were consensus scored by 2 scorers. (Graph Presented) When 1 prospective observation exceeded the baseline our algorithm detected 229 of 237 (97%) ASDA arousals with 68 false positives. When 3 prospective points exceeded the filter threshold it detected 198 of 237(84%) ASDA arousals with 19 false positives. Neural network EEG analysis can be post processed to detect some ASDA arousals. It also detects other events which may represent arousal from sleep not detected using the ASDA criteria. |
| spellingShingle | Bennett, L Stradling, JR Barbour, C Davies, R Detection of cortical EEG arousals from a Neural Network (NN) analysis |
| title | Detection of cortical EEG arousals from a Neural Network (NN) analysis |
| title_full | Detection of cortical EEG arousals from a Neural Network (NN) analysis |
| title_fullStr | Detection of cortical EEG arousals from a Neural Network (NN) analysis |
| title_full_unstemmed | Detection of cortical EEG arousals from a Neural Network (NN) analysis |
| title_short | Detection of cortical EEG arousals from a Neural Network (NN) analysis |
| title_sort | detection of cortical eeg arousals from a neural network nn analysis |
| work_keys_str_mv | AT bennettl detectionofcorticaleegarousalsfromaneuralnetworknnanalysis AT stradlingjr detectionofcorticaleegarousalsfromaneuralnetworknnanalysis AT barbourc detectionofcorticaleegarousalsfromaneuralnetworknnanalysis AT daviesr detectionofcorticaleegarousalsfromaneuralnetworknnanalysis |