Development of Chinese food picture library for inducing food cravings

Abstract

Cue-induced food cravings are strong desires directed toward specific foods, usually ones with high caloric content, and can lead to overeating. However, although food cravings vary according to individual preferences for specific high-calorie food subtypes, a structured library of food craving-inducing pictures including multiple categories of high-calorie foods does not yet exist. Here, we developed and validated a picture library of Chinese foods (PLCF) consisting of five subtypes of high-calorie foods (i.e., sweets, starches, salty foods, fatty foods, and sugary drinks) to allow for more nuanced future investigations in food craving research, particularly in Chinese cultural contexts. We collected 100 food images representing these five subtypes, with four food items per subtype depicted in five high-resolution photographs each. We recruited 241 individuals with overweight or obesity to rate the food pictures based on craving, familiarity, valence, and arousal dimensions. Of these participants, 213 reported the severity of problematic eating behaviors as a clinical characteristic. Under the condition of mixing multiple subtypes of high-calorie foods, we did not observe significant differences in craving ratings for high- and low-calorie food images (p _tukey > 0.05). Then, we compared each subtype of high-calorie food images to low-calorie ones, and found craving ratings were greater for the images of salty foods and sugary drinks (ps < 0.05). Furthermore, we conducted a subgroup analysis of individuals according to whether they did or did not meet the criteria for food addiction (FA) and found that greater cravings induced by the images of high-calorie food subtypes (i.e., salty foods and sugary drinks) only appeared in the subgroup that met the FA criteria. The results show that the PLCF is practical for investigating food cravings.

Keywords: cue-induced craving; food addiction; food craving; obesity; salty foods.

FIGURE 1

The development process of the Picture Library of Chinese Foods (PLCF). From left to right, this figure describes the steps to construct the PLCF. Step 1, nomination: we first defined subtypes of foods with reference to the YFAS 2.0, invited local people with obesity to nominate their favorite foods, and finally defined image search terms based on foods that were frequently nominated and fell into the predefined subtypes. Step 2, search: we searched the Internet for and downloaded publicly licensed images of the most frequently nominated foods. The inclusion criteria were: (1) resolution greater than 1,024 × 768 pixels; (2) the food was presented in the center of the image within a natural scene; and (3) no special social information, such as human faces or hands, appeared. Step 3, structure: the classification hierarchy of the PLCF was finalized. The picture library contains five food subtypes, each with four items depicted in five different images. Step 4, measures: the picture set was divided into ten subsets according to a planned missing data design. Each subset included twenty high-calorie images without food item repetitions and one low-calorie and one spoiled food image as the internal references. Each rater rated only one subset. More details are described in the section “Materials and methods”.

FIGURE 2

Statistical analysis of PLCF ratings. (A) Histogram of one-way repeated-measures ANOVA results for each rating dimension. All ratings of palatable food (i.e., high- and low-calorie) images were significantly higher than those of spoiled food. Ratings of high- and low-calorie images significantly differed on the valence and arousal dimensions but not on the craving dimension. Solid bars indicate high-calorie foods, hollow bars indicate low-calorie foods, and diagonal lines indicate spoiled foods [same in panels (A–C)]. (B) Scatter chart of the Pearson correlation between craving ratings and craving experience frequency in daily life. The frequency of craving experiences in life was positively correlated with the cravings induced by high-calorie images and was uncorrelated with those induced by low-calorie images. A linear fit was conducted separately, and the shading within the dotted lines indicates the 95% CI. (C) Dot plot of the results from one-way repeated-measures ANOVA with six levels (i.e., five subtypes of high-calorie food and low-calorie food). The cravings induced by salty food and sugary drink images were greater than those induced by low-calorie images. The diamond symbols indicate subtypes; the circles indicate food types. (D) Heatmap of the correlation coefficient between the craving ratings of high-calorie food images and rater characteristics. The cravings induced by high-calorie food images were positively correlated with the baseline desire to eat and the severity of food addiction measured by YFAS 2.0. However, there was only one correlation with the severity of eating disorder symptoms measured by EDE-Q. EDE-Q, eating disorder examination questionnaire; YFAS, Yale Food Addiction Scale 2.0. For all subplots, *** indicates p < 0.001, ** indicates p < 0.01, * indicates p < 0.05, and “ns” indicates no significance. The error bar indicates the 95% CI.

FIGURE 3

Statistical analysis of PLCF data stratified by FA criteria. (A) Histogram of independent samples T-test results for each rating dimension. Between subgroups that did and did not meet FA criteria, ratings of high-calorie images showed significant differences in the craving, valence, and arousal dimensions but not in the familiarity dimension. The colors indicate different subgroups, with red denoting the subgroup that met FA criteria and blue denoting the subgroup that did not meet FA criteria [same in panels (A–C)]. (B) Scatter chart of the Pearson correlation between craving ratings of high-calorie food images and the frequency of cravings in daily life. The positive correlation between craving ratings of high-calorie food images and daily craving experiences (shown in Figure 2B) only appeared in the subgroup that met FA criteria. A linear fit was conducted separately, and the shading within the dotted line indicates the 95% CI. (C) Dot plot of the estimated marginal means from the two-way repeated-measures ANOVA with the food subtype as a within-subject factor (six levels) and the FA classification as a between-subject factor (two levels). The greater cravings induced by salty food and sugary drink images (compared to low-calorie food images, shown in Figure 2C) only appeared in the subgroup that met FA criteria. For all subplots, *** indicates p < 0.001, ** indicates p < 0.01, and “ns” indicates no significance. The error bar indicates the 95% CI.