AI-powered dynamic scattering liquid crystal biosensor for label-free detection of ovarian cancer biomarker CA125

Conventional static liquid crystal (LC) biosensors are frequently constrained by limited optical contrast and insufficient sensitivity for trace biomarker detection. Here, we report a label-free biosensing platform based on dual-frequency liquid crystals (DFLCs) operated in the dynamic scattering mode (DSM). By driving the DFLC system at the crossover frequency of 22.2 kHz and an optimized voltage of 22.5 V, it entered an electrohydrodynamically unstable state, where minute interfacial perturbations induced by biomolecular recognition were amplified into high-contrast macroscopic scattering responses. Using haze as a quantitative readout, the DSM optical response stabilized within 3 s of electric-field application, and the platform achieved a limit of detection (LOD) of 0.11 ng/mL for bovine serum albumin, 0.82 and 4.86 ng/mL for cancer antigen 125 (CA125) in deionized water and human serum, respectively, demonstrating high sensitivity and robust resistance to matrix interference. To enable automated quantitative analysis and concentration prediction from the highly complex DSM optical textures, we implemented a ResNet-34-based convolutional neural network (CNN), which utilizes a residual backbone for hierarchical feature extraction from polarized optical microscopy images, integrated with a custom regression head. The deep learning model achieved a cross-validation coefficient of determination R² of 0.84, and 82% of the independent blind-evaluation samples were correctly assigned to their experimentally prepared concentration groups. The integration of DSM-based signal amplification with AI-driven quantitative analysis positions the DFLC biosensor as a rapid, sensitive, and automated analytical platform with strong potential for next-generation in vitro diagnostics.