Computed tomography (CT) examinations pose challenges for continuous patient observation, particularly when adverse events such as severe reactions to contrast media occur. To improve patient monitoring in such situations, this preliminary study proposes a contactless method for respiratory waveform estimation using a depth camera and AI-based body detection. The method identifies anatomically relevant respiratory regions and extracts depth-based motion signals while subjects are seated facing the camera, which is positioned approximately 2 m away. Performance was evaluated experimentally using a wearable force-sensor respiration belt as the reference. Quantitative assessment was conducted using waveform error metrics, Pearson correlation coefficients, respiratory-rate agreement, and Bland–Altman analysis, while qualitative analysis was used to examine the influence of clothing conditions on measurement performance. The results show that the proposed method can provide stable respiratory waveform estimation, with the chest region yielding the lowest waveform error and the highest correlation among the evaluated ROIs. Bland–Altman analysis further indicated small systematic errors in respiratory-rate estimation, although variability-related indices were affected by ROI selection and clothing conditions. These findings support the feasibility of the proposed approach for contactless respiratory monitoring during CT examinations and indicate that the main contribution of this study is to clarify the importance of anatomical ROI selection for robust waveform extraction under CT-oriented monitoring conditions.
Kojima et al. (Mon,) studied this question.