PPG based Noninvasive Blood Glucose Monitoring using Multi-view Attention and Cascaded BiLSTM Hierarchical Feature Fusion Approach.

Diabetes is a chronic disease with exponential growth and poses significant challenges to global healthcare. Regular blood glucose (BG) monitoring is key for avoiding diabetic complications. Traditional BG measurement techniques are invasive and minimally invasive, causing pain, discomfort, cost, and infection risks. To address these issues, we developed a noninvasive BG monitoring approach on photoplethysmography (PPG) signals using multi-view attention and cascaded BiLSTM hierarchical feature fusion approach. Firstly, we implemented a convolutional multi-view attention block to extract the temporal features through adaptive contextual information aggregation. Secondly, we built a cascaded BiLSTM network to efficiently extract the fine-grained features through bidirectional learning. Finally, we developed a hierarchical feature fusion with bilinear polling through cross-layer interaction to obtain higher-order features for BG monitoring. For validation, we conducted comprehensive experimentation on up to 6 days of PPG and BG data from 21 participants. The proposed approach showed competitive results compared to existing approaches by RMSE of 1.67 mmol/L and MARD of 17.88%. Additionally, the clinical accuracy using Clarke error grid (CEG) analysis showed 98.80% of BG values in Zone A+B. Therefore, the proposed approach offers a favorable solution in diabetes management by noninvasively monitoring the BG levels.

Automatic Chinese Food recognition based on a stacking fusion model.

With commercialization of deep learning models, daily precision dietary record based on images from smartphones becomes possible. This study took advantage of Deep-learning techniques on visual recognition tasks and proposed a big-data-driven Deep-learning model regressing from food images. We established the largest data set of Chinese dishes to date, named CNFOOD-241. It contained more than 190,000 images with 241 categories, covering Staple food, meat, vegetarian diet, mixed meat and vegetables, soups, dessert category. This study also compares the prediction results of three popular deep learning models on this dataset, ResNeXt101_32x32d achieving up to 82.05% for top-1 accuracy and 97.13% for top-5 accuracy. Besides, this paper uses a multi-model fusion method based on stacking in the field of food recognition for the first time. We built a meta-learner after the base model to integrate the three base models of different architectures to improve robustness. The accuracy achieves 82.88% for top-1 accuracy.Clinical Relevance-This study proves that the application of artificial intelligence technology in the identification of Chinese dishes is feasible, which can play a positive role in people who need to control their diet, such as diabetes and obesity.