Accuracy of an Artificial Intelligence–Based Model for Estimating Leftover Liquid Food in Hospitals: Validation Study
Bibliographic record
Abstract
Background An accurate evaluation of the nutritional status of malnourished hospitalized patients at a higher risk of complications, such as frailty or disability, is crucial. Visual methods of estimating food intake are popular for evaluating the nutritional status in clinical environments. However, from the perspective of accurate measurement, such methods are unreliable. Objective The accuracy of estimating leftover liquid food in hospitals using an artificial intelligence (AI)–based model was compared to that of visual estimation. Methods The accuracy of the AI-based model (AI estimation) was compared to that of the visual estimation method for thin rice gruel as staple food and fermented milk and peach juice as side dishes. A total of 576 images of liquid food (432 images of thin rice gruel, 72 of fermented milk, and 72 of peach juice) were used. The mean absolute error, root mean squared error, and coefficient of determination (R2) were used as metrics for determining the accuracy of the evaluation process. Welch t test and the confusion matrix were used to examine the difference of mean absolute error between AI and visual estimation. Results The mean absolute errors obtained through the AI estimation approach were 0.63 for fermented milk, 0.25 for peach juice, and 0.85 for the total. These were significantly smaller than those obtained using the visual estimation approach, which were 1.40 (P<.001) for fermented milk, 0.90 (P<.001) for peach juice, and 1.03 (P=.009) for the total. By contrast, the mean absolute error for thin rice gruel obtained using the AI estimation method (0.99) did not differ significantly from that obtained using visual estimation (0.99). The confusion matrix for thin rice gruel showed variation in the distribution of errors, indicating that the errors in the AI estimation were biased toward the case of many leftovers. The mean squared error for all liquid foods tended to be smaller for the AI estimation than for the visual estimation. Additionally, the coefficient of determination (R2) for fermented milk and peach juice tended to be larger for the AI estimation than for the visual estimation, and the R2 value for the total was equal in terms of accuracy between the AI and visual estimations. Conclusions The AI estimation approach achieved a smaller mean absolute error and root mean squared error and a larger coefficient of determination (R2) than the visual estimation approach for the side dishes. Additionally, the AI estimation approach achieved a smaller mean absolute error and root mean squared error compared to the visual estimation method, and the coefficient of determination (R2) was similar to that of the visual estimation method for the total. AI estimation measures liquid food intake in hospitals more precisely than visual estimation, but its accuracy in estimating staple food leftovers requires improvement.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.006 | 0.020 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".