用于伤口感染早期诊断的电子鼻动态温度调制传感系统和智能算法研究

Trauma has become a big public hazard in modern society, and wound infection is the most common complication of trauma. It is very significant for clinical practice to research a quick and efficient diagnosis technology of wound infection. Electronic nose (E-nose)can diagnose the infection by analyzing the headspace odor of wound, and the existing research shows its effectiveness. But in the early stage of wound infection, the concentration of metabolic product is extremely low, and the performance of E-nose in the early diagnosis of wound infection is not ideal. This project will adopt temperature modulated sensing technology to realize the early diagnosis of wound infection. The working temperature of sensors is not stationary but changed over time, and there will be more useful information which can represent different pathogenic bacteria and this can finally improve the lowest concentration detection limit of E-nose. In addition, the samples of wound infection are difficult to obtain, and the sample size of the training set is insufficient, which is a further obstacle for limiting the improvement of the recognition rate. This project focuses on sensor array construction, temperature modulation model setting, data feature extraction and fusion, and training sample shortage solutions based on transfer learning and semi-supervised learning. In order to establish the early diagnosis model of human trauma based on E-nose, this model can diagnose the infection of wound as early as possible, as soon as possible and accurately. The research of this project can promote the theoretical development of sensor working temperature modulation technology and wound diagnosis technology of E-nose, and promote the clinical progress of E-nose technology.

创伤已成为社会一大公害，而伤口感染又是创伤的最常见并发症，快速、高效的诊断在临床实践中具有重要意义。电子鼻技术通过分析伤口顶空气体成分进行感染诊断，已有的研究已表明其有效性。但在伤口感染早期，病原菌代谢产物浓度极低，导致电子鼻的诊断效果并不理想。本项目将温度调制传感技术用于伤口感染的早期诊断，在传感器响应中包含更多的可表征不同病原菌的有用信息，从而提高电子鼻的最低浓度检测限。另外伤口感染样本难获得，导致训练电子鼻的样本量不足，这是限制识别率提升的又一障碍，本项目围绕传感器阵列构建、温度调制模型设置、数据特征提取和融合以及基于迁移学习和半监督学习的训练样本不足解决方案等展开研究，以求建立基于电子鼻的人体创伤早期诊断模型，该模型可尽早、尽快、准确地诊断出伤口的感染情况。本项目的研究可推进传感器工作温度调制技术和电子鼻伤口诊断技术的理论发展，并促进电子鼻技术的临床进程。

创伤已成为社会一大公害，而伤口感染又是创伤的最常见并发症，快速、高效的诊断在临床实践中具有重要意义。电子鼻技术通过分析伤口顶空气体成分进行感染诊断，已有的研究已表明其有效性。但在伤口感染早期，病原菌代谢产物浓度极低，导致电子鼻的诊断效果并不理想。本项目将温度调制传感技术用于伤口感染的早期诊断，从而提高电子鼻的最低浓度检测限。另外借助迁移学习和半监督学习解决伤口样本小的问题。本项目严格按照计划进行，构建出了用于伤口感染检测的温度调制传感阵列，并提出了一系列的智能嗅觉算法用于提升系统的诊断正确率，这对于电子鼻的临床化和理论研究都有重要意义，本项目执行过程中共发表SCI/EI期刊论文13篇，申请发明专利5项，其中授权3项，参加学术交流会议3次，获得了完整的硬件系统设计方案、伤口感染嗅觉数据集和算法代码库。