Department of Health Technology and Informatics, Hong Kong Polytechnic University, Kowloon, Hong Kong.
Department of Ultrasound, Fifth Affiliated Hospital of Sun Yat-sen University, Zhuhai, P.R. China.
Ren Fail. 2023 Dec;45(1):2202755. doi: 10.1080/0886022X.2023.2202755.
Given its progressive deterioration in the clinical course, noninvasive assessment and risk stratification for the severity of renal fibrosis in chronic kidney disease (CKD) are required. We aimed to develop and validate an end-to-end multilayer perceptron (MLP) model for assessing renal fibrosis in CKD patients based on real-time two-dimensional shear wave elastography (2D-SWE) and clinical variables.
From April 2019 to December 2021, a total of 162 patients with CKD who underwent a kidney biopsy and 2D-SWE examination were included in this single-center, cross-sectional, and prospective clinical study. 2D-SWE was performed to measure the right renal cortex stiffness, and the corresponding elastic values were recorded. Patients were categorized into two groups according to their histopathological results: mild and moderate-severe renal fibrosis. The patients were randomly divided into a training cohort ( = 114) or a test cohort ( = 48). The MLP classifier using a machine learning algorithm was used to construct a diagnostic model incorporating elastic values with clinical features. Discrimination, calibration, and clinical utility were used to appraise the performance of the established MLP model in the training and test sets, respectively.
The developed MLP model demonstrated good calibration and discrimination in both the training [area under the receiver operating characteristic curve (AUC) = 0.93; 95% confidence interval (CI) = 0.88 to 0.98] and test cohorts [AUC = 0.86; 95% CI = 0.75 to 0.97]. A decision curve analysis and a clinical impact curve also showed that the MLP model had a positive clinical impact and relatively few negative effects.
The proposed MLP model exhibited the satisfactory performance in identifying the individualized risk of moderate-severe renal fibrosis in patients with CKD, which is potentially helpful for clinical management and treatment decision-making.
鉴于其在临床病程中的逐渐恶化，需要对慢性肾脏病（CKD）患者的肾纤维化严重程度进行非侵入性评估和风险分层。我们旨在开发和验证一种基于实时二维剪切波弹性成像（2D-SWE）和临床变量的 CKD 患者肾纤维化评估的端到端多层感知器（MLP）模型。
2019 年 4 月至 2021 年 12 月，本单中心、横断面、前瞻性临床研究共纳入 162 例接受肾活检和 2D-SWE 检查的 CKD 患者。进行 2D-SWE 以测量右肾皮质硬度，并记录相应的弹性值。根据组织病理学结果将患者分为两组：轻度和中重度肾纤维化。将患者随机分为训练队列（n=114）或测试队列（n=48）。使用机器学习算法的 MLP 分类器构建了一个诊断模型，该模型将弹性值与临床特征相结合。在训练和测试集中，分别使用判别、校准和临床实用性来评估所建立的 MLP 模型的性能。
所开发的 MLP 模型在训练组[接受者操作特征曲线下的面积（AUC）=0.93；95%置信区间（CI）=0.88 至 0.98]和测试组[AUC=0.86；95%CI=0.75 至 0.97]中均表现出良好的校准和判别能力。决策曲线分析和临床影响曲线也表明，MLP 模型具有积极的临床影响和相对较少的负面影响。
所提出的 MLP 模型在识别 CKD 患者中重度肾纤维化的个体风险方面表现出令人满意的性能，这可能有助于临床管理和治疗决策。
