Usability and accuracy of an artificial intelligence-based automatic meal tray photography device for estimating liquid food intake in healthcare settings.

BACKGROUND & AIMS: Malnutrition affects hospital care and disease treatment. The evaluation of food intake is essential for assessing the nutritional status of patients. We developed a meal tray photography device that can automatically capture pictures of leftover food and upload them to an artificial intelligence model for analysis to improve the usability of food intake estimation. METHODS: In this study, the usability of food intake estimation using the automatic meal tray photography device and the tablet device was evaluated using the System Usability Scale (SUS). The accuracies of the automatic meal tray photography device, tablet device, and visual estimation method were compared with that of the weighing method as the benchmark. RESULTS: The mean SUS score of the automatic meal tray photography device was 63.2 (out of 100), which was significantly higher than that of 56.0 for the tablet device and higher than that of the tablet device for all questions. However, the accuracy of the automatic meal tray photography device tended to be lower than that of the tablet device, and the visual estimation method had the highest accuracy. CONCLUSION: This study is valuable as it can help reduce the pressure on staff in healthcare settings.