Estimation of Meal Residues Using Foundation Models

Hideaki Masumoto, Tatsuhito Hasegawa, Atsuharu Horiuchi, Yuta Nakano, Kaisei Wada, Junko Yotsuya

Abstract: In the medical field, to ascertain the nutrition ingested by patients, the amount of meal residues from meals served in the hospital is recorded. However, the recording of the amount of meal residues is done entirely manually, placing a burden on nurses. In this study, we propose an automatic recording system for the amount of meal residues in nursing tasks, using the foundational models Grounding DINO and Segment Anything in High Quality (HQ-SAM). The validation results show that the proposed method can estimate the amount of meal residues in zero-shot from the meal regions.