Reliability of noninvasive hyperspectral tongue diagnosis for menstrual diseases using machine learning method.

The outward appearance of human tongue can reflect changes in blood circulation caused by pathological states, and it has been used as an assisted method for clinical diseases diagnosis for thousands of years in China. The purpose of this study is to observe the changes in the tongue of patients with menstrual-related diseases in hyperspectral imaging and to explore the value of hyperspectral tongue imaging combining with machine learning algorithm (HSI-ML) in the diagnosis of menstrual diseases (MD). Hyperspectral tongue images are collected from 429 patients with five different kinds of MD and 52 participants with normal menstruation. Tongue coating and tongue body spectral characteristics are extracted and used as model input variables to investigate the influence on the modeling results.Normalization (Norm), first derivative (1st D), second derivative (2nd D), savitzky-golay smoothing (S-G), multiplicative scatter correction (MSC), and standard normal variate transformation (SNV) are used as preprocessing method. Four model algorithms, k-nearest neighbor (KNN), random forest (RF), support vector machines (SVM) and artificial neural network(ANN) are used and compared. Experimental results show that patients of each MD group exhibit significantly lower spectral reflectance of tongue coating and tongue body (P < 0.05) than participants of normal menstruation group. And the modeling results indicate that the "2nd D + S-G + ANN" identification model based on tongue body spectral characteristics yields the optimal performance. Specifically, its accuracy, macro-precision, macro-recall, and macro-F1 score are 0.9729, 0.9697, 0.9703, and 0.97, respectively. It indicates that HSI-ML method with hyperspectral tongue images can provide a rapid and noninvasive detection method for MD screening.