A rapid approach for discriminating Ganoderma species using attenuated total reflectance–Fourier transform infrared (ATR-FTIR) spectroscopy integrated with chemometric analysis and convolutional neural network (CNN)

The issue of adulteration and misclassification of <i>Ganoderma</i> species is addressed in this research. In the study, we present a novel and comprehensive framework for <i>Ganoderma</i> authentication by analyzing attenuated total reflectance-Fourier transform infrared (ATR-FTIR) spectra using a combined approach of a chemometric analysis and deep learning (DL) with a convolutional neural network (CNN). The three <i>Ganoderma</i> species involved in this study were as follows: <i>Ganoderma lucidum</i>, <i>Ganoderma sinense</i>, and <i>Ganoderma tsugae</i>. Among chemometric models, orthogonal partial least squares discriminant analysis (OPLS-DA) yielded a high accuracy of 98.61%, a sensitivity of 97.92%, and a specificity of 98.96%. Additionally, the root-mean-squared error of estimation (RMSEE), root-mean-squared error of prediction (RMSEP), and root-mean-squared error of cross-validation (RMSECV) values for the OPLS-DA model were <0.3, confirming its reliability. The CNN model also performed well, achieving 89.84% accuracy, 84.75% sensitivity, and 92.38% specificity, with minimal variation during random segregation testing. Additionally, the model exhibited a precision of 0.87 ± 0.02, a recall of 0.85 ± 0.03, and an F1 score of 0.86 ± 0.03 for 10 random segregation tests. As a conclusion, both chemometric and CNN models developed in this study are efficient and robust for classifying <i>Ganoderma</i> species. To further validate this combined approach, we aim to implement chemometric and CNN models in other medicinal herb authentication in the future.