基于图像识别的智能营养追踪患者膳食特征分析

Dietary Characteristic Analysis of Patients Using Intelligent Image-Based Nutrition Tracking

摘要: 患者膳食特征的精准分析是营养干预的关键。为克服传统方法主观性强、数据不精确等局限，本研究构建了一种基于图像识别的智能营养追踪系统。系统通过移动设备采集膳食图像，集成深度学习技术实现食物识别与分割，结合参照物标定估算食物重量，并基于标准数据库计算营养素摄入量。以上海市嘉定区中心医院2025年9月30日至11月30日期间11名慢性肾脏病规律透析患者上传的158张膳食图像(对应197条食物记录)为样本进行分析。统计检验结果显示，三餐间能量摄入存在显著差异(F(2,25) = 3.387, p = 0.050)，午餐能量(1626.0 ± 948.4 kcal)显著高于早餐(708.1 ± 520.0 kcal)；蛋白质摄入同样存在显著差异(H = 7.872, p = 0.020)，午餐蛋白质(90.4 ± 53.1 g)显著高于早餐(29.4 ± 21.3 g)。用户平均每餐摄入3种食物，高频食物包括米饭、牛奶、吐司面包等。早餐食物种类(均值2.4种)较午餐(3.6种)与晚餐(3.2种)更为单一；个体膳食多样性差异显著，日摄入种类在12种以内。研究表明，该群体膳食结构呈现午餐能量与蛋白质双高模式，餐次间差异具有统计学意义。本研究为基于图像的膳食评估提供了实证依据。

Abstract: Accurate analysis of patient dietary characteristics is crucial for nutritional intervention. To overcome the limitations of traditional methods, such as subjectivity and imprecise data, this study developed an intelligent nutrition tracking system based on image recognition. The system captures dietary images via mobile devices, integrates deep learning technology for food identification and segmentation, estimates food weight using reference object calibration, and calculates nutrient intake based on a standard database. A total of 158 dietary images (corresponding to 197 food records) uploaded by 11 chronic kidney disease patients undergoing regular dialysis at Shanghai Jiading District Central Hospital between September 30 and November 30, 2025, were analyzed. Statistical tests revealed significant differences in energy intake among the three meals (F(2,25) = 3.387, p = 0.050), with lunch energy (1626.0 ± 948.4 kcal) significantly higher than breakfast (708.1 ± 520.0 kcal). Protein intake also showed significant differences (H = 7.872, p = 0.020), with lunch protein (90.4 ± 53.1 g) significantly higher than breakfast (29.4 ± 21.3 g). Users consumed an average of 3 types of food per meal, with high-frequency foods including rice, milk, and toast bread. Breakfast food variety (mean 2.4 types) was simpler compared to lunch (3.6 types) and dinner (3.2 types). Individual dietary diversity varied significantly, with daily intake ranging up to 12 food types. The study indicates that this group exhibits a “high energy and protein intake at lunch” dietary pattern, with statistically significant differences among meals. This research provides empirical evidence for image-based dietary assessment.