This study presents a novel non-destructive method for assessing walnut internal quality using Low-Field Magnetic Resonance Imaging (LF-MRI) and radiomics technology. Due to the hard shell of walnuts, determining their interior quality is challenging. By analyzing walnut kernel Low-Field Nuclear Magnetic Resonance (LF-NMR) relaxation curve characteristics and LF-MRI imaging, radiomics techniques were employed to extract, select, and reduce the dimensionality of features from MRI images. Ten significant features strongly correlated with walnut kernel rancidity were identified, and machine learning models were built using six optimized classification algorithms. The Random Forest (RF) models achieved impressive performance with a test accuracy of 93.52%, test recall scores of 92.78%, and test F1 scores of 96.81%. Additionally, the RF model demonstrated a higher net benefit within the threshold probability range of 0.02 to 0.98, as indicated by the DCA curve. The study's findings have significant implications for the walnut industry and food quality control, providing a reliable and efficient means of detecting and identifying walnut kernel rancidity.
Read full abstract