Deep learning for label-free characterization of nanoscale particles in interferometric microscopy

Label-free interferometric microscopy enables detection of nanoscale particles below the diffraction limit, but particle characterization is fundamentally limited because interferometric contrast depends on particle polarizability. Since polarizability depends jointly on particle size and refractive index, different particles can produce similar focal-plane intensities, preventing reliable identification in heterogeneous samples. We present a depth-resolved characterization approach that combines axial interferometric signatures with hierarchical deep learning. A common-path interferometric microscope using a Si/SiO₂ layered substrate was used to acquire defocus image stacks by translating the sample along the optical axis. Axial scanning modulates the phase between the reference reflection and scattered field, producing a particle-specific intensity profile. Individual particle signals were extracted using preprocessing adapted from the Depth Scanning Correlation method and analyzed using a two-stage 1D convolutional neural network. Using polystyrene and silica nanoparticles, the classifier achieved ~92% material classification accuracy and distinguished particles with similar focal-plane contrast. The results demonstrate that full axial interferometric signatures provide discriminative information beyond single-plane measurements, enabling automated, label-free characterization of heterogeneous nanoparticle populations.