2023 Fiscal Year Final Research Report
Automatic Classification of Neonatal Sleep-Wake States by Video Analysis
| Project/Area Number |
21K12704
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 90130:Medical systems-related
|
|---|
| Research Institution | Mie University |
|---|
Principal Investigator |
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
盛田 健人 三重大学, 工学研究科, 准教授 (40844626)
新小田 春美 福岡女学院看護大学, 看護学部, 教授 (70187558)
|
|---|
| Project Period (FY) |
2021-04-01 – 2024-03-31
|
| Keywords | パターン認識 / 医用画像処理 / 動画像解析 / 深層学習 |
|---|
| Outline of Final Research Achievements |
In this study, we proposed a method for automatically classifying the sleep-wake state of a child in a NICU in a non-contact manner using video. We compared a machine learning method using motion features obtained from optical flow and a deep learning method using 3DResNet for both videos showing the whole body and videos from which face regions were extracted. Furthermore, the results of the classification method using 3DResNet for videos with face regions extracted and the method using 3DResNet for videos of the entire body were integrated based on output probability after time series smoothing, yielding a classification accuracy of 0.611 and a Kappa score of 0.623.
|
| Free Research Field |
知覚情報処理
|
| Academic Significance and Societal Importance of the Research Achievements |
機器によるバイタルデータ（眼球電位，筋電位，脳波，呼吸等）の測定により正確な睡眠覚醒状態を評価できることが報告されているが，新生児の行動観察に基づくNBASのStateとは一致せず，新生児への負担が大きいため長時間の連続測定は難しい．また，これまでにBrazeltonのNBASに基づく睡眠覚醒状態を継続的・客観的に自動分類する手法は存在しなかった．
本研究の成果により新生児や看護師に負担をかけずに継続的な観測が可能になり，看護師の主観に左右されない客観的な睡眠覚醒状態の調査ができるため，NICUの明暗環境と新生児睡眠覚醒状態の関係の大規模調査を可能にするという点で意義がある．
|
