| 研究課題/領域番号 |
21K17670
|
|---|
| 研究種目 |
若手研究
|
| 配分区分 | 基金 |
|---|
| 審査区分 |
小区分59040:栄養学および健康科学関連
|
|---|
| 研究機関 | 京都先端科学大学 |
|---|
研究代表者 |
梁 滋路 京都先端科学大学, 工学部, 講師 (10782807)
|
|---|
| 研究期間 (年度) |
2021-04-01 – 2025-03-31
|
| 研究課題ステータス |
交付 (2023年度)
|
| 配分額 *注記 |
4,680千円 (直接経費: 3,600千円、間接経費: 1,080千円)
2024年度: 650千円 (直接経費: 500千円、間接経費: 150千円)
2023年度: 520千円 (直接経費: 400千円、間接経費: 120千円)
2022年度: 1,560千円 (直接経費: 1,200千円、間接経費: 360千円)
2021年度: 1,950千円 (直接経費: 1,500千円、間接経費: 450千円)
|
|---|
| キーワード | sleep apnea / SpO2 / ensemble learning / machine learning / multiscale entropy / oximeter / wearable computing / ubiquitous computing / personal informatics / digital health / mHealth / chaos theory |
|---|
| 研究開始時の研究の概要 |
SAHS screening markers will be identified through nonlinear analysis of the chaotic dynamics of the physiological data measurable by wearable devices. These markers, together with features derived using data science techniques, will be used to develop SAHS screening algorithms.
|
| 研究実績の概要 |
The achievement this year revolved around enhancing the performance of SpO2-based sleep apnea screening models. A new feature set was constructed leveraging machine learning and chaos analysis techniques. Probabilistic ensemble approach was applied to construct apnea screening models at three cutoff points: 5, 15, 30 events/h. These models underwent thorough evaluation using multiple performance metrics and rigorous statistical analysis. The impact of decision boundaries and data granularity were systematically explored, marking the first comprehensive investigation of its kind in machine learning based sleep apnea screening. The models outperform existing models by a substantial margin. The findings have led to several publications in international journals and conferences, indexed in Web of Science and Scopus.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
The project progressed as planned during this fiscal year. A new feature set was constructed by combining classic machine learning and chaos analysis techniques. Utilizing this feature set, probabilistic ensemble models were constructed using three tuned and calibrated base classifiers: SVM, logistic regression, and light gradient boosting machine. Model performance was evaluated using multiple measures, considering varying decision boundaries and data granularity. The developed models demonstrated superior performance across all three AHI cutoffs compared to existing sleep apnea screening models. The new approach allows for elegant integration of the pre-test sleep apnea prevalence into model tuning, thereby enhancing its clinical relevance and applicability in real-world scenarios.
|
| 今後の研究の推進方策 |
We have identified two key directions for future research. Firstly, we plan to explore the effectiveness of an original feature engineering method, which involves multiscale feature extraction. This approach aims to capture more nuanced patterns and relationships within the data, potentially enhancing the performance of our sleep apnea screening models. Secondly, we intend to conduct a comprehensive assessment of the external validity and generalizability of the developed sleep apnea screening models. This evaluation will involve testing the models on other datasets to ensure their robustness and applicability across different contexts. By addressing these areas in future work, we aim to further advance the effectiveness and reliability of our models in real-world settings.
|
