乳児の授乳時腸蠕動音解析に基づく腸音特徴空間の有効特徴量抽出を用いた1日の総おむつ替え回数予測

抄録

In 2024, the total fertility rate in Japan declined to 1.15, with annual births falling below 700,000 for the first time. The need to prepare for and manage diaper changes while outside the home poses a psychological burden on parents. In this study, we aimed to develop a noninvasive method to record infant bowel sounds at the onset of feeding session and predict the total number of daily diaper changes using XGBoost with a limited set of features. Fourteen healthy infants (seven boys and seven girls, aged 2-11 months) participated in the study, which was approved by the Ethics Committee of Ehime University Hospital. Bowel sounds were recorded for 10 minutes immediately after feeding session began using a small device attached inside the diaper. Our method consists of the following steps: (Step 0) Noninvasive bowel sound recording and classification into six characteristic types based on duration and maximum frequency (MF) under physician supervision; (Step 1) Calculation of the bowel sound ratio by dividing the number of occurrences of each characteristic sound by the total number of sounds and normalizing the values; (Step 2) Clustering of the feature space using bowel sound ratio and MF to derive four-dimensional effective features (4D) ;and (Step 3) Classification of the number of daily diaper changes (6-10 times/day, five classes) as the target variable using XGBoost, with SMOTE applied to correct for class imbalance. The comparison conditions included 6D, 6D＋SMOTE, and 4D, with performance evaluated using accuracy and F1-score. Statistical significance was assessed using a t-test (p-value) and Cohen’s d effect size. The proposed method (4D＋SMOTE) achieved an accuracy of 0.86 and an F1-score of 0.82, with a significant difference observed between 6D and 4D＋SMOTE (t＝5.78, p＝0.0045, dz＝2.58 for F1-score) . Moreover, since features with a low proportion of bowel sounds contributed to prediction performance, the findings suggest that bowel sound analysis may serve as an effective indicator of diaper change frequency. This approach enables accurate estimation of daily diaper changes, facilitates outing planning, and may help prevent social isolation and reduce parenting-related stress.