【摘要】 职业性慢性铅中毒是一种逐渐发展的病症，因早期症状不典型而忽略致使铅累积并发展为重度铅中毒，严重威胁着职业群体的身体健康及生活质量。尽管现有技术(如原子荧光光谱)已应用于现场即时监测，但其复杂的设备结构和分析时间限制了快速原位检测的普及。激光诱导击穿光谱(LIBS)作为一种新兴的检测手段在元素分析领域展现出巨大的潜力和应用前景。采用LIBS技术结合机器学习算法，证明了职业性慢性铅中毒类别快速诊断的可行性。优化了全血样品制样方法，提出超声处理可使全血基质分布更加均匀从而缓解激光作用于干燥血液时的样品碎裂问题。相比滤纸、石墨和硼酸基板，载玻片是最合适的基板类型。探讨了不同LIBS实验参数对铅元素特征谱线信号强度与信背比的影响。采集了不同类型职业性慢性铅中毒的模拟血液LIBS数据，利用主成分分析(PCA)提取特征实现数据降维。采用10折交叉验证和支持向量机(SVM)、反向传播神经网络(BPNN)构建慢性铅中毒类别诊断模型。两种模型的识别准确度均达到90%以上，为低浓度铅元素光谱弱，传统方法难区分问题提出了解决方案，并且BPNN模型表现出优异的诊断效果，分类准确率和精确率分别为95.56%和96.08%。结果表明：基于机器学习的LIBS技术可实现全血中铅元素超标的及时筛检，为快速准确筛查血铅异常值提供了辅助检测方法，为职业性慢性铅中毒的诊断补充了临床依据。更多还原

【Abstract】 Occupational chronic lead poisoning is a gradually developing disease.It is easy to accumulate lead and develop into severe lead poisoning due to atypical early symptoms, which seriously threaten the health and quality of life of occupational groups. Mainstream analytical techniques have problems, such as requiring high precision but complicated and time-consuming operations, or being easy to operate but having poor applicability, which cannot achieve rapid in-situ detection. Laser-induced breakdown spectroscopy(LIBS), a new detection method, has demonstrated great potential and promising applications in the field of elemental analysis. In this paper, the feasibility of rapidly diagnosing occupational chronic lead poisoning is demonstrated using LIBS technology combined with machine learning algorithms. The whole blood sample preparation method has been optimized, and it is proposed that ultrasonic treatment can make the whole blood matrix more uniform, thereby alleviating the sample fragmentation problem when the laser is applied to dry blood. A glass slide is the most suitable substrate type compared to filter paper, graphite, and boric acid substrates. The effects of various LIBS experimental parameters on the signal intensity and signal-to-background ratio of the characteristic spectral lines of the element lead(Pb) were investigated. Simulated blood LIBS data for different types of occupational chronic lead poisoning were collected, and data dimensionality reduction was achieved by extracting features using principal component analysis(PCA). A 10-fold cross-validation and support vector machine(SVM) and back-propagation neural network(BPNN) were used to construct a diagnostic model for the chronic lead poisoning category. The recognition accuracies of both models reached more than 90%, which solved the problem of weak spectral difference of Pb elements in low concentration, which was difficult to be distinguished by traditional methods, and the BPNN model showed excellent diagnostic effect, with the classification accuracy and precision rate of 95.56% and 96.08%, respectively. The results demonstrate that machine learning-based LIBS technology can facilitate the timely detection of lead elemental excess in whole blood, providing an auxiliary method for the rapid and accurate screening of blood lead abnormalities, and supplementing the clinical basis for the diagnosis of occupational chronic lead poisoning.更多还原