基于显微高光谱成像技术的尿沉渣结晶样本分析

The analysis of urine components, called urine sediment, is paramount in clinical practice. By observing particles, cells, and crystals in urine sediment, doctors can obtain important information about the patient's urogenital health, which is crucial for diagnosing urogenital-related ailments. However, identifying urine sediment crystals heavily relies on medical professionals' manual observation under a microscope, which is time-consuming, subjective, and often inaccurate. Consequently, automated microscopic urine sediment image analysis using image analysis technology has gained significant attention. However, these methods rely solely on morphological information to classify crystal samples, making distinguishing between morphology-similar crystals difficult, resulting in low classification accuracy. Microscopic hyperspectral imaging technology integrates spatial and spectral information, revealing distinct spectral characteristics as different substances exhibit varying degrees of light absorption and scattering across different spectral bands. In this study, we introduced microscopic hyperspectral imaging technology to analyze urine sediment crystal samples and used a self-developed microscopic hyperspectral imaging system to acquire hyperspectral images. We collected microscopic hyperspectral image data of five urine sediment crystal sample types, including calcium oxalate, cystine, calcium phosphate, uric acid, and triple phosphate. Additionally, we trained four machine learning classifiers support vector machine (SVM) , k-nearest neighbors (KNN) , decision tree (DT) , and neural network(NN) models on this dataset to classify the five types of urine sediment crystal samples. The classification accuracies of SVM, KNN, DT, and NN models for the five types of urine sediment crystals reached 0. 959 8, 0. 959 8, 0. 982 9, and 0. 991 7, respectively. Our research indicates that applying microscopic hyperspectral imaging technology to urine sediment sample analysis enables the acquisition of spatial information and facilitates the extraction of discriminative spectral features, thereby assisting physicians in microscopic examination and supporting the popularization of urine sediment microscopy techniques.