Label-free classification of nanoscale drug delivery systems using hyperspectral imaging and convolutional neural networks

Label-free characterization of nanoscale drug delivery systems remains a critical challenge in pharmaceutical research. Traditional analytical methods, such as cryo-electron microscopy, are labor-intensive, low-throughput, and often require labeling, which can interfere with nanoparticle functionality. This study introduces a non-invasive hyperspectral imaging (HSI) framework combined with deep learning to classify therapeutic liposomes. A 3D convolutional neural network (3D CNN) was employed to extract spatial-spectral features, while the synthetic minority oversampling technique (SMOTE) addressed class imbalance common in pharmaceutical datasets. Control and doxorubicin-loaded liposomes were imaged using dark-field HSI (VNIR 400-1000 nm). Dimensionality reduction (PCA), patch extraction, and SMOTE were applied before training the 3D CNN model. Model performance was evaluated using overall accuracy, F1-score, and Cohen's Kappa metrics. The proposed 3D CNN-SMOTE model achieved a classification accuracy of 99.16% with near-perfect F1-scores across all classes. This label-free HSI framework enables robust, scalable classification of liposomal drug carriers, offering a promising tool for real-time, non-destructive quality control during nanoparticle formulation and manufacturing. This approach broadly applies to pharmaceutical development, including batch verification, encapsulation efficiency screening, and regulatory compliance workflows.