Quantifying the effect of caloric and non-caloric sweeteners in the brain response using EEG and convolutional neural network

Sweetener type can influence sensory properties and consumer’s acceptance and preference for low-calorie products. An ideal sweetener does not exist, and each sweetener must be used in situations to which it is best suited. Aspartame and sucralose can be good substitutes for sucrose in passion fruit juice. Despite the interest in artificial sweeteners, little is known about how artificial sweeteners are processed in the human brain. Here, we evaluated brain signals of 11 healthy subjects when they tasted passion fruit juice equivalently sweetened with sucrose (9.4 g/100 g), sucralose (0.01593 g/100 g), or aspartame (0.05477 g/100 g). Electroencephalograms were recorded for two sites in the gustatory cortex (i.e., C3 and C4). Data with artifacts were disregarded, and the artifact-free data were used to feed a CNN. Our results indicated that the brain responses distinguish juice sweetened with different sweeteners with an average accuracy of 0.823. Practical Applications Finding sweeteners that best fit consumer preferences evolves understanding how the gustatory cortex processes sweeteners. Ideal equivalence will occur when the brain is no longer able to distinguish stimuli that are consciously perceived. This study presents a method of signal acquisition using a single channel and an open-source processing environment. This would allow, for example, to disregard the use of a commercial electroencephalograph and expand the studies in this area and offering to food industry additional tools in the development of products sweetened with non-caloric sweeteners.


INTRODUCTION 40
Sugar has been the main sweetener in human diet for centuries. It represents a high 41 percentage of the human daily energy consumption, but it offers little additional nutritional 42 value (Bassoli & Merlini, 2003;Caballero, 2013;Zorn et al., 2014). 43 Given that replacing sugar with non-caloric (or low-calorie) sweeteners have become 44 popular among consumers seeking to lose or to maintain weight. Moreover, foods with the same 45 sweetening capacity might be perceived differently due to their caloric content (Frank et al., Any study involving equi-intense sweeteners (i.e., sweeteners used in equivalent 57 amounts) must consider how sweeteners influence sensory properties and consumer's 58 acceptance and preference for low-calorie products (Pinheiro et al., 2005). An ideal sweetener 59 does not exist, so each sweetener is appropriate for specific situations (Nabors, 2002). Passion 60 fruit is a popular tropical fruit that has an important commercial variety named yellow passion 61 fruit, which is used to prepare juice that requires sweetening (Deliza et al., 2005). Aspartame 62 and sucralose can be good substitutes for sucrose in passion fruit juice (Rocha & Bolini, 2015a, 63 2015b. 64 layer is usually derived from the combination of one or more planes of previous layers. The 92 nodes of a plane are connected to a small region of each connected plane of the previous layer. 93 Each node of the convolution layer extracts the features from the input by convolution 94 operations on the input nodes. As the features propagate to the highest layer or level, the 95 dimensions of features are reduced depending on the kernel size for the convolutional and max-96 pooling operations, respectively. The output of the last CNN layer is used as the input for a fully 97 connected network, which is called classification layer. In the classification layer, the extracted 98

MATERIAL AND METHODS 117
The methods used herein were divided into three main sections: (1) Participant selection, (2) 118 acquisition of EEG signal from the selected group, and (3) signal processing and CNN 119 (Convolutional Neural Network) training and tests. Our aim was to use a CNN to determine 120 differences between stimuli. 121

Stimuli 122
The methods used herein were divided into three main sections: (1) Participant selection, (2) 123 acquisition of EEG signal from the selected group, and (3) signal processing and CNN 124 (Convolutional Neural Network) training and tests. Our aim was to use a CNN to determine 125 differences between stimuli. 126 127

Participant Selection 128
A total of 105 volunteers were included in this study. The volunteers comprised students, 129 teachers, and employees aged 19-55 years, recruited on campus. They did not have diabetes, 130 smoke, or use medications that affect taste or cognitive processes. Preference was given to 131 volunteers that were used to consuming passion fruit juice (or that at least had no aversion to the 132 fruit taste). 133 Each volunteer received samples containing 30 mL of passion fruit juice sweetened with 134 different amounts of sugar (i.e., 4.7, 7.05, 9.4, 11.75, and 14.1 g). The samples were placed in 135 disposable cups and numbered randomly between 0000 and 9999 (e.g., A7932). The 136 participants received samples in a randomized sequence of concentrations and had to answer the 137 following question: "How much sugar did you have in your juice?" Answers were collected in a 138 9.0-cm visual scale (VAS). VASs are input mechanisms that allow users to specify a value 139 within a predefined range. The volunteers were instructed to consider the centre of the scale as 140 ideal sweetness, 0 as less sweet than ideal, and 9 as sweeter than ideal. Our objective was to 141 and, among these individuals, to select those that had good ability to order the samples 143 according to the sugar concentration. 144 The volunteers were informed about the nature and aims of the experiments and provided 145 informed consent. The study was approved by the Ethics Committee of the Animal Science and The last network layer contained four neurons, which were responsible for mapping 195 four possible results of an entry (water, sugar, aspartame, or sucralose). A categorical cross-196 entropy loss function was used.

Participant selection 213
sucrose concentration of 9.4 g/100 g and good ability to order the sample concentrations. This 215 means that the selected participants indicated values around 4.5 cm on the VAS, which was 216 equivalent to a sucrose concentration of 9.4 g/100 g. Furthermore, they were able to order all the 217 concentrations properly. Of the individuals considered fit, 11 agreed to participate in the brain 218 signal acquisition stage. 219 sugar concentration of 9.4 g/100 g as his favourite. He was selected for the next step. 222 Participants B and C were not selected. Participant B (Figure 1-c) placed the samples correctly, 223 but his preferred sample was not the sample with sugar concentration of 9.4 g/100 g. Although 224 participant C (Figure 1-d) preferred the sample with a sugar concentration of 9.4 g/100 g, he 225 was not able to order the samples correctly. 226 227 Figure 2 shows the network architecture that achieved the best performance. The best 229 number for n was 20, as can be seen by the input layer shape. The other three fully connected 230 layers had 16, 16, and 64 neurons. Its dropout optimal value was 0.2. 231

Signal processing and CNN training 228
We used this optimal architecture to train the CNN with 70 % of the data, while we 232 employed the remaining 30% for the test. 233 For visualization, Figure 3 presents the confusion matrix for this dataset. In the 234 confusion matrix, the horizontal axis is the predicted label, and the vertical axis is the true label. 235 The elements on the diagonal represent the numbers of correctly classified samples. 236 237 whereas 35 samples (7.7%) were incorrectly predicted as aspartame and 4 samples (2.6%) were 239 incorrectly predicted as sucralose. For the sucrose class, 31 samples (20.0%) were correctly 240 predicted as sucrose, whilst 25 samples (16.1%) were incorrectly predicted as aspartame, and 91 241 samples (58.7%) were incorrectly predicted as sucralose. As for the aspartame class, 119 242 samples (64.0%) were correctly classified, while 67 samples (36.0%) were incorrectly predicted 243 as sucralose. Concerning the sucralose class, 186 samples (100.0%) were correctly classified. 244 The main difficulty lay in the classes of water and sucrose. 245 Table 1 lists the metrics results for the four stimuli (classes). 247 The average metrics (Table 1)  When we consider reference (water) only, the classification in Figure 3 indicated that the 252 average identification accuracy is 91,4% compared to an overall classification of 82.3%. This 253 work uses a drink instead of a solution and, in a first study of this nature, it was important to be 254 able to distinguish a critical reference well. 255 When we compare general performance with reference performance, mainly f1 score that 256 consider recall and precision measures simultaneously we can deduce that misclassifications in 257 sweeteners classes are greater than that in reference class. 258 Analysing Figure 3 in more detail, suggests that the network is more sensitive to the low-calorie 259 sweetener classes than to the sucrose class. By evaluating the false positives (FP) of the 260 classification, aspartame can be predicted to be sucralose, but not sugar. In turn, sucralose had 261 no false positives. The most surprising result is that sucrose can be classified both as sugar and 262 low-calorie sweetener. In other words, evaluated low-calorie sweeteners had not been confused 263 with the caloric sweetener, but the caloric sweetener can be confused with the low-calorie 264 sweetener. Andersen et al. (2019) observed that similar tastes that are consciously 265 indistinguishable can result in different brain cortical activations. A similar result was obtained 266 when gustatory evoked potentials (GEPs) were used to assess the brain response to sucrose, 267 aspartame, and stevia in humans (Mouillot et al., 2020). The authors stated that, although 268 sucrose, aspartame, and stevia led to the same taste perception, the GEPs showed that cerebral 269 activation by these different sweet solutions had different recordings. They suggested that,

CONCLUSION 302
We have compared brain signals acquired in response to the consumption of passion fruit juice 303 sweetened with sucrose (caloric sweetener), sucralose, or aspartame (low-calorie sweeteners). 304 We used the artifact-free data to feed a CNN. The results indicated that the brain responses can 305 distinguish the juice sweetened with sucrose from the juice sweetened with aspartame and 306 sucralose.  We also thank Ajinomoto for the donation of aspartame (AminoSweet TM ). 312