Evaluation of dietary intake in healthcare facilities is crucial for targeted and timely nutritional and health intervention. Current classical methods do not allow for an objective automated approach to dietary intake measurements. This Master's thesis explores the feasibility of automated, cost-effective optical volumetric analysis based on 3D-profilometry and assesses measurement and method errors. Our successful prototype demonstrates that the key step of the method, estimating mass from a 3D-image, can be fully automated for foods compatible with the system. Based on the analysis of the actual menu at UKC Ljubljana, it was found that 93% of the meals are already compatible with the method. To achieve full feasibility, 7% of the meals would need to be adjusted. The measurement error of the system is smaller than comparable systems from the literature (CV(m) = 6.6%; MAPE(m) = 9.9%) and is acceptable even after normalization based on the actual hospital menu (CV(B, M, OH, E) ≈ 9.7%). The primitive method of segmentation based on colour profiles requires upgrading or replacement with a better method to reduce systemic errors. Suggestions for further improvements were provided. The method has the potential to be transferred into practice and significantly improve the evaluation of dietary intake in healthcare facilities. No major technological obstacles to feasibility were identified. The main barrier to practical implementation is the substantial capital investment required for the development of a production system.
|
