Iodinated X-ray contrast media (ICM) are commonly detected at considerable concentrations in aquatic environments. The long-term pollution trends in ICM at the whole lake/river scale have not yet been investigated; therefore, the risks associated with ICM and the influences of environmental factors remain understudied. Herein, the occurrence and distribution of ICM in the surface water of Taihu Lake were comprehensively investigated. In addition, the accuracy and applicability of different machine learning models for predicting ICM pollution and associated health risk were compared using meteorological and water quality parameters as inputs. The results revealed that the Σ₇ICM concentration ranged from 10.8 to 454.6 ng/L, exhibiting significant spatial and seasonal variations, which reflected the influence of hydrodynamics and climatic conditions. Among the nine models, the RF model achieved the most accurate prediction of ICM, with R² ≥ 0.92. Via feature importance ranking and linear relationship analysis, TN, NH₃-N, S_275–295, PS, SUVA₂₅₄, UV₂₅₄, and pH were identified as important factors affecting ICM. This study provides a hybrid framework that includes environmental pollution prediction, health risk analysis, and key environmental factor identification for ICM, providing scientific techniques for the application of machine learning in the analysis of trace organic contaminants.

中文翻译：

---

利用机器学习准确预测太湖碘化 X 射线造影剂的污染和健康风险，揭示关键环境因素


碘 X 射线造影剂 （ICM） 通常在水生环境中以相当高的浓度检测到。尚未调查整个湖泊/河流规模的 ICM 的长期污染趋势;因此，与 ICM 相关的风险和环境因素的影响仍未得到充分研究。本文对太湖表层水中 ICM 的赋存和分布进行了全面调查。此外，使用气象和水质参数作为输入，比较了不同机器学习模型预测 ICM 污染和相关健康风险的准确性和适用性。结果表明，Σ₇ICM 浓度范围为 10.8 至 454.6 ng/L，表现出显著的空间和季节变化，反映了流体动力学和气候条件的影响。在这 9 个模型中，RF 模型实现了对 ICM 最准确的预测，R² ≥ 0.92。通过特征重要性排序和线性关系分析，确定 TN 、 NH_3-N 、 S_275–295 、 PS 、 SUVA₂₅₄ 、 UV₂₅₄ 和 pH 值是影响 ICM 的重要因素。本研究提供了一个混合框架，包括 ICM 的环境污染预测、健康风险分析和关键环境因素识别，为机器学习在痕量有机污染物分析中的应用提供了科学技术。