Janicki Joseph J, Zwaans Bernadette M M, Bartolone Sarah N, Ward Elijah P, Chancellor Michael B
Underactive Bladder Foundation, Pittsburgh, PA 15235, USA.
Corewell Health William Beaumont University Hospital, Royal Oak, MI 48073, USA.
Diagnostics (Basel). 2024 Dec 5;14(23):2734. doi: 10.3390/diagnostics14232734.
This study aimed to improve machine learning models for diagnosing interstitial cystitis/bladder pain syndrome (IC/BPS) by comparing classical machine learning methods with newer AutoML approaches, utilizing biomarker data and patient-reported outcomes as features. We applied various machine learning techniques to biomarker data from the previous IP4IC and ICRS studies to predict the presence of IC/BPS, a disorder impacting the urinary bladder. Data were sourced from two nationwide, crowd-sourced collections of urine samples involving 2009 participants. The models utilized included logistic regression, support vector machines, random forests, k-nearest neighbors, and AutoGluon. Expanding the dataset for model training and evaluation resulted in improved performance metrics compared to previously published findings. The implementation of AutoML methods yielded enhancements in model accuracy over classical techniques. The top-performing models achieved a receiver-operating characteristic area under the curve (ROC-AUC) of up to 0.96. This research demonstrates an improvement in model performance relative to earlier studies, with the top model for binary classification incorporating objective urinary biomarker levels. These advancements represent a significant step toward developing a reliable classification model for the diagnosis of IC/BPS.
本研究旨在通过比较经典机器学习方法与更新的自动机器学习（AutoML）方法，利用生物标志物数据和患者报告的结果作为特征，改进用于诊断间质性膀胱炎/膀胱疼痛综合征（IC/BPS）的机器学习模型。我们将各种机器学习技术应用于先前IP4IC和ICRS研究中的生物标志物数据，以预测IC/BPS的存在，IC/BPS是一种影响膀胱的疾病。数据来自两个全国性的、众包的尿液样本收集，涉及2009名参与者。所使用的模型包括逻辑回归、支持向量机、随机森林、k近邻和AutoGluon。与先前发表的研究结果相比，扩大用于模型训练和评估的数据集导致性能指标得到改善。AutoML方法的实施使模型准确性相对于经典技术有所提高。表现最佳的模型实现了高达0.96的曲线下受试者工作特征面积（ROC-AUC）。这项研究表明，相对于早期研究，模型性能有所提高，二元分类的最佳模型纳入了客观的尿液生物标志物水平。这些进展代表了朝着开发用于诊断IC/BPS的可靠分类模型迈出的重要一步。
