Answers are grounded in this paper and cite their sources.
Conventional surface plasmon resonance imaging (SPRi) relies on relative resonance-shift measurements and laborious multipoint calibration, rendering it incompatible with disposable-chip and point-of-care scenarios. Here, we report a Hadamard-encoded spectral compressive reconstruction (HSCR) platform coupled with a lightweight one-dimensional convolutional neural network (1D-SPRNet). By reframing chip-scale spatial nonuniformity as an in situ data-augmentation mechanism, the system generates ∼20,000 pixel-wise training spectra from a single measurement cycle and converts six compressed CMOS frames into quantitative refractive-index changes (ΔRIU) using only six multiplexed LED measurements. By outputting physical refractive-index units rather than arbitrary resonance shifts, the system eliminates the need for exhaustive chip-specific calibration: a minimal two-point Langmuir fit suffices for quantitative immunodetection. We demonstrate high-fidelity spectral recovery (Pearson correlation >0.999) and cross-chip generalization using NaCl refractive-index standards. In a rabbit IgG/goat antirabbit IgG model, the pretrained 1D-SPRNet achieves linear regression R2 = 0.996 in phosphate-buffered saline and R2 = 0.987 in 10-fold diluted human serum, with relative errors within ±7% in phosphate-buffered saline and recoveries ranging from 88.7% to 101.1% in 10-fold diluted human serum, along with negligible nonspecific adsorption. This integration of compressive spectral encoding with end-to-end deep regression enables high-throughput, real-time, minimal-calibration multiplexed biosensing.
No takes yet. Share an insight, caveat, or question.
Zeng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: