Objective Raman spectroscopy is proposed as a next-generation method for the identification of monosodium urate (MSU) and calcium pyrophosphate (CPP) crystals in synovial fluid. As the interpretation of Raman spectra requires specific expertise, the method is not directly applicable for clinicians. We developed an approach to demonstrate that the identification process can be automated with the use of machine learning techniques. The developed system is tested in a point-of-care-setting at our outpatient rheumatology department. Methods We collected synovial fluid samples from 446 patients with various rheumatic diseases from three centres. We analysed all samples with our Raman spectroscope and used 246 samples for training and 200 samples for validation. Trained observers classified every Raman spectrum as MSU, CPP or other. We designed two one-against-all classifiers, one for MSU and one for CPP. These classifiers consisted of a principal component analysis model followed by a support vector machine. Results The accuracy for classification of CPP using the 2023 ACR/EULAR CPPD classification criteria was 96.0% (95% CI: 92.3, 98.3), while the accuracy for classification of MSU using the 2015 ACR/EULAR gout classification criteria was 92.5% (95% CI: 87.9, 95.7). Overall, the accuracy for classification of pathological crystals was 88.0% (95% CI: 82.7, 92.2). The model was able to discriminate between pathological crystals, artifacts and other particles such as microplastics. Conclusion We here demonstrate that potentially complex Raman spectra from clinical patient samples can be successfully classified by a machine learning approach, resulting in an objective diagnosis independent of the opinion of the medical examiner.

中文翻译：

---

通过在床旁环境中获取的拉曼光谱的机器学习，客观诊断痛风和焦磷酸钙沉积病


目的 拉曼光谱是鉴定滑液中尿酸单钠 （MSU） 和焦磷酸钙 （CPP） 晶体的下一代方法。由于拉曼光谱的解析需要特定的专业知识，因此该方法并不直接适用于临床医生。我们开发了一种方法来证明识别过程可以通过使用机器学习技术实现自动化。开发的系统在我们门诊风湿病科的护理点环境中进行了测试。方法 我们从 3 个中心的 446 例各种风湿病患者收集滑液样本。我们使用拉曼光谱仪分析了所有样品，并使用 246 个样品进行训练，使用 200 个样品进行验证。训练有素的观察者将每个拉曼光谱分为 MSU、CPP 或其他。我们设计了两个一对一的分类器，一个用于 MSU，一个用于 CPP。这些分类器由一个主成分分析模型和一个支持向量机组成。结果 使用 2023 年 ACR/EULAR CPPD 分类标准对 CPP 进行分类的准确性为 96.0% （95% CI： 92.3， 98.3），而使用 2015 年 ACR/EULAR 痛风分类标准对 MSU 进行分类的准确性为 92.5% （95% CI： 87.9， 95.7）。总体而言，病理晶体分类的准确性为 88.0% （95% CI： 82.7， 92.2）。该模型能够区分病理晶体、伪影和其他颗粒，如微塑料。结论我们在这里证明，来自临床患者样本的潜在复杂拉曼光谱可以通过机器学习方法成功分类，从而得到独立于法医意见的客观诊断。