Huang Jurong, Ding Hang, McBride Simon, Ireland David, Karunanithi Mohan
The Australian e-Health Research Centre, CSIRO, Royal Brisbane and Women's Hospital, Brisbane, Australia.
Stud Health Technol Inform. 2015;214:121-7.
Over 380 million adults worldwide are currently living with diabetes and the number has been projected to reach 590 million by 2035. Uncontrolled diabetes often lead to complications, disability, and early death. In the management of diabetes, dietary intervention to control carbohydrate intake is essential to help manage daily blood glucose level within a recommended range. The intervention traditionally relies on a self-report to estimate carbohydrate intake through a paper based diary. The traditional approach is known to be inaccurate, inconvenient, and resource intensive. Additionally, patients often require a long term of learning or training to achieve a certain level of accuracy and reliability. To address these issues, we propose a design of a smartphone application that automatically estimates carbohydrate intake from food images. The application uses imaging processing techniques to classify food type, estimate food volume, and accordingly calculate the amount of carbohydrates. To examine the proof of concept, a small fruit database was created to train a classification algorithm implemented in the application. Consequently, a set of fruit photos (n=6) from a real smartphone were applied to evaluate the accuracy of the carbohydrate estimation. This study demonstrates the potential to use smartphones to improve dietary intervention, although further studies are needed to improve the accuracy, and extend the capability of the smartphone application to analyse broader food contents.
全球目前有超过3.8亿成年人患有糖尿病，预计到2035年这一数字将达到5.9亿。糖尿病若未得到控制，常常会引发并发症、导致残疾并造成过早死亡。在糖尿病管理中，通过饮食干预控制碳水化合物摄入量对于将每日血糖水平控制在推荐范围内至关重要。传统的干预方式依靠自我报告，通过纸质日记来估算碳水化合物摄入量。众所周知，传统方法不准确、不方便且资源消耗大。此外，患者通常需要长期学习或培训才能达到一定的准确性和可靠性。为解决这些问题，我们提出设计一款智能手机应用程序，该程序能根据食物图像自动估算碳水化合物摄入量。该应用程序利用图像处理技术对食物类型进行分类、估算食物量，并据此计算碳水化合物的含量。为验证概念，创建了一个小型水果数据库来训练应用程序中实现的分类算法。随后，使用一组来自真实智能手机的水果照片（n = 6）来评估碳水化合物估算的准确性。本研究证明了使用智能手机改善饮食干预的潜力，不过还需要进一步研究来提高准确性，并扩展智能手机应用程序分析更广泛食物成分的能力。
