Academic Department of Biomedicine and Prevention, University of Rome Tor Vergata, and Clinical Department of Surgical Sciences, Section of Gynecology, Tor Vergata University Hospital, Viale Oxford, 81 - 00133, Rome, Italy.
Department of Electronic Engineering, University of Rome Tor Vergata, Via del Politecnico, 1 - 00133, Rome, Italy.
Sci Rep. 2020 May 14;10(1):7970. doi: 10.1038/s41598-020-64512-4.
RPL is a very debated condition, in which many issues concerning definition, etiological factors to investigate or therapies to apply are still controversial. ML could help clinicians to reach an objectiveness in RPL classification and access to care. Our aim was to stratify RPL patients in different risk classes by applying an ML algorithm, through a diagnostic work-up to validate it for the appropriate prognosis and potential therapeutic approach. 734 patients were enrolled and divided into 4 risk classes, according to the numbers of miscarriages. ML method, called Support Vector Machine (SVM), was used to analyze data. Using the whole set of 43 features and the set of the most informative 18 features we obtained comparable results: respectively 81.86 ± 0.35% and 81.71 ± 0.37% Unbalanced Accuracy. Applying the same method, introducing the only features recommended by ESHRE, a correct classification was obtained only in 58.52 ± 0.58%. ML approach could provide a Support Decision System tool to stratify RPL patients and address them objectively to the proper clinical management.
RPL 是一种极具争议的病症，其定义、病因研究因素以及治疗方法仍存在诸多争议。ML 可以帮助临床医生在 RPL 分类和获得治疗方面实现客观化。我们的目的是通过诊断工作来应用 ML 算法将 RPL 患者分层为不同的风险类别，以验证其进行适当预后和潜在治疗方法的有效性。共纳入 734 例患者，并根据流产次数将其分为 4 个风险类别。使用称为支持向量机（SVM）的 ML 方法对数据进行分析。使用全部 43 个特征和信息量最大的 18 个特征集，我们获得了可比的结果：分别为 81.86±0.35%和 81.71±0.37%不平衡准确率。应用相同的方法，仅引入 ESHRE 推荐的特征，正确分类率仅为 58.52±0.58%。ML 方法可以提供支持决策系统工具，将 RPL 患者分层，并客观地将其分配到适当的临床管理中。
