Alternation of inverse problem approach and deep learning for lens-free microscopy image reconstruction

A lens-free microscope performs in-line holographic measurements without focusing optics; image formation is recovered computationally by solving an inverse problem. Gradient-based optimization suffers from local minima and phase wrapping errors when the optical thickness exceeds λ/2, which commonly occurs for cells in suspension and during mitosis, limiting live-cell applications. We propose an alternation approach that combines inverse-problem optimization with deep learning. A first reconstruction generates an initial, phase-wrapped estimate that is fed to a convolutional neural network trained to correct phase wrapping. The network output initializes a second reconstruction that enforces data fidelity and mitigates CNN prediction errors. We demonstrate that this approach resolves phase wrapping for dense suspensions of cells, a regime where conventional inverse optimization alone fails.