A groundbreaking study has unveiled a rapid, non-destructive method to assess the quality of Ganoderma lucidum—a prized medicinal mushroom—using hyperspectral imaging (HSI) coupled with advanced machine learning algorithms. The new approach promises to revolutionize quality control in the traditional Chinese medicine industry by significantly reducing the time and cost associated with conventional chemical testing.

Ganoderma lucidum, renowned for its antioxidant, antibacterial, anti-inflammatory, and tumor-inhibiting properties, has long been used in healthcare and nutrition. However, its quality can vary widely depending on cultivation conditions, harvest times, and geographical origin. Traditional testing methods such as high-performance liquid chromatography (HPLC) and nuclear magnetic resonance (NMR) spectroscopy, while accurate, are often time-consuming, destructive, and require complex sample preparation.

In response to these challenges, the research team turned to hyperspectral imaging—a technique that captures both spatial and spectral data across a wide range of wavelengths. By scanning samples in the visible near-infrared (VNIR) and short-wave infrared (SWIR) bands, the researchers were able to collect comprehensive data on the chemical composition of both Ganoderma caps (GLC) and their powdered forms (GLP). The study focused on two key compounds: polysaccharides, known for their immune-boosting effects, and ergosterol, a precursor to vitamin D.

To transform the wealth of spectral data into meaningful predictions, the team employed three machine learning models: a backpropagation neural network (BPNN), a decision tree (DT), and an extreme learning machine (ELM). Among these, the ELM model—enhanced through a genetic algorithm (GA) that incorporated a voting-based feature selection method—stood out, delivering outstanding predictive accuracies with determination coefficients (R² values) reaching 0.96 for polysaccharides and 0.97 for ergosterol.

The innovation didn’t stop there. The researchers also experimented with advanced data processing techniques, such as principal component analysis (PCA) and iterative versus voting-based genetic algorithm methods, to extract the most relevant spectral wavelengths. The voting-based GA, which aggregates multiple independent GA runs, proved particularly robust by reducing randomness in wavelength selection and enhancing overall model stability.

Interestingly, the study found that the intact Ganoderma cap yielded more reliable predictions than its powdered counterpart. This outcome is attributed to the fact that pulverization, while increasing sample uniformity, can disrupt the mushroom’s natural structure and potentially affect the chemical interactions critical for accurate spectral analysis.

Beyond its application to Ganoderma lucidum, this research highlights the vast potential of combining HSI with machine learning for non-invasive quality assessment. Industries ranging from agriculture to food safety and pharmaceutical production could benefit from the ability to quickly and accurately evaluate product quality without damaging the sample.

This pioneering work not only sets a new standard for rapid quality testing in traditional medicine but also opens the door for broader applications of hyperspectral imaging technology. By streamlining the evaluation process, the method could facilitate more consistent quality control, boost consumer confidence, and drive further innovation in the fields of food and medicinal product safety.

（Source: Frontiers）