3. Synovial Fluid Density-Based Non-Invasive Detection of Knee Osteoarthritis Using Bioelectrical Impedance and LSTM Networks
基于滑液密度的生物电阻抗与LSTM网络膝骨关节炎无创检测方法
Dr. Ulhas V. Patil and Dr. Bhushankumar Narayan Shinde
如果用于学术论文、证书或正式出版物，通常建议保留英文姓名，并在首次出现时附中文音译。
Abstract: Knee osteoarthritis (KOA) is a progressive degenerative joint disorder that requires early and reliable diagnosis to prevent severe functional impairment. Conventional imaging techniques mainly detect structural damage at later stages and often involve high cost or radiation exposure. This study proposes a synovial fluid density–based non-invasive diagnostic framework using bioelectrical impedance analysis combined with a Long Short-Term Memory (LSTM) deep learning model. A low-amplitude alternating current is injected into the knee joint region to capture impedance variations influenced by synovial fluid composition, viscosity, and ionic concentration. Extracted impedance features such as resistance, reactance, magnitude, and phase angle are used to estimate synovial fluid density and generate temporal feature sequences. These sequences are processed by an LSTM network to classify osteoarthritis severity into normal, mild, and severe categories. Experimental results demonstrate stable training convergence with high classification performance, achieving improved accuracy, precision, recall, and F1-score. The proposed method provides a low-cost, portable, and radiation-free solution for early detection and continuous monitoring of knee osteoarthritis, showing strong potential for real-time clinical decision support.
摘要: 膝骨关节炎（Knee Osteoarthritis，KOA）是一种进行性退行性关节疾病，需要进行早期且可靠的诊断，以防止严重的功能障碍。传统影像学技术主要用于检测疾病后期出现的结构性损伤，并且通常存在成本较高或具有辐射暴露等问题。本研究提出了一种基于滑液密度的无创诊断框架，采用生物电阻抗分析（Bioelectrical Impedance Analysis，BIA）结合长短期记忆网络（Long Short-Term Memory，LSTM）深度学习模型，实现膝骨关节炎的早期检测与严重程度分类。
该方法向膝关节区域施加低幅值交流电流，通过采集受滑液成分、黏度及离子浓度影响的阻抗变化，获取相关生物电阻抗信号。进一步提取电阻、抗抗、阻抗幅值和相位角等阻抗特征，用于估计滑液密度并构建时序特征序列。随后，将这些时序特征序列输入LSTM网络，对膝骨关节炎的严重程度进行分类，将其划分为正常、轻度和重度三个类别。
实验结果表明，该模型具有稳定的训练收敛性能，并表现出较高的分类性能，在准确率、精确率、召回率和F1值等指标方面均取得了较好的结果。所提出的方法具有低成本、便携式和无辐射等优势，可为膝骨关节炎的早期检测及持续监测提供一种新的技术方案，并在实时临床决策支持方面展现出较大的应用潜力。计。
