Cancer diagnosis method based on multi-spectral diffraction imaging for cell recognition

To overcome drawbacks such as the need for biomarker labels, low automation, low-throughput, low integration, and high cost, an online cancer cell recognition application based on extracted multi-spectral lens-free cell diffraction fingerprint features is proposed. A high-throughput lens-free wide-field diffraction imaging platform captures diffraction images of four cancer and two normal cell types under multi-spectral illumination. Cell diffraction fingerprints are normalized by binary masking, features are extracted via Gray Level Co-occurrence Matrix (GLCM), and a Bayesian-Optimized Support Vector Machine (SVM) performs model training and recognition. Multi-spectral recognition accuracy is compared with five single-spectral modes. Co-culture experiments of cancer and normal cells are designed to validate the recognition accuracy of the models. This engineering application enables accurate real-time recognition of four types of cancer cells and two types of normal cells with accuracies of 96.0%, 98.0%, 95.2%, and 96.0% for HeLa, Huh-7, A549, and MCF-7 cells, respectively. The core advantages of this application lie in its complete elimination of chemical labeling or staining and the achievement of rapid imaging and recognition using a portable device, showing potential in reducing operational costs, simplifying workflows, and facilitating clinical application.