| Project/Area Number |
22K21186
|
|---|
| Research Category |
Grant-in-Aid for Research Activity Start-up
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
0908:Society medicine, nursing, and related fields
|
|---|
| Research Institution | Fujita Health University |
|---|
Principal Investigator |
He Yupeng 藤田医科大学, 医学部, 助教 (00953267)
|
|---|
| Project Period (FY) |
2022-08-31 – 2024-03-31
|
| Project Status |
Completed (Fiscal Year 2023)
|
| Budget Amount *help |
¥1,300,000 (Direct Cost: ¥1,000,000、Indirect Cost: ¥300,000)
Fiscal Year 2023: ¥650,000 (Direct Cost: ¥500,000、Indirect Cost: ¥150,000)
Fiscal Year 2022: ¥650,000 (Direct Cost: ¥500,000、Indirect Cost: ¥150,000)
|
|---|
| Keywords | Artificial image / 人工画像 / 疫学研究 / 予測モデル / 機械学習 / Model performance / Prediction / Epidemiology / Feature sequence / Artifical image / Prediction method / 1 |
|---|
| Outline of Research at the Start |
This study consists of artificial images generation and optimal pattern verification. Medical data are transformed as pixel values to reversely generate artificial images by series of patterns. Labeling the images with the target disease, the processed dataset is divided into training set and test set. The classifier (trained by training set) which has the highest accuracy (tested by test set) is the optimal prediction model, and the corresponding pattern is the optimal pattern. The optimal pattern is used to generate images to realize the visualization of the medical data.
|
| Outline of Final Research Achievements |
A novel method using artificial image was developed to enhance the model precision in epidemiology study. The concept was inspired from image identification. Pixels in digital images are used as features when training the identification model. The order-related relationship is assumed to exist in epidemiological data. Given a certain dataset, features are transformed to pixel values for generating artificial images. Orders of pixels are randomly permutated and the model is trained using pixel-permutated artificial image sample sets. In the preliminary experiment, one binary response was designed to be predicted by 76 features. 10,000 artificial image sample sets were randomly selected for model training. Models’ performance (area under the receiver operating characteristic curve values) depicted a bell-shaped distribution. Namely, feature order information had a strong impact on model performance. Our novel model construction strategy has potential to enhance model predictability.
|
| Academic Significance and Societal Importance of the Research Achievements |
従来の疫学研究でよく使われる線形モデルと比較して、本研究で開発した新手法は、特徴を2次元人工画像の形式で配置することで、１）モデルの精度を向上させる。２）複数の特徴間の交絡要因を究明できる。３）ブラックボックスのような機械学習モデルを視覚的に説明できる。４）特徴の位置を使用して特徴の重要性を説明する。５）疫学調査以外の順序不特定のデータの分析に活用できる。
|
