Accelerated Wirtinger Flow for Multiplexed Fourier Ptychographic\n Microscopy

Emrah Bostan, Mahdi Soltanolkotabi, David Ren, Laura Waller

Abstract

Fourier ptychographic microscopy enables gigapixel-scale imaging, with both large field-of-view and high resolution. Using a set of low-resolution images that are recorded under varying illumination angles, the goal is to computationally reconstruct high-resolution phase and amplitude images. To increase temporal resolution, one may use multiplexed measurements where the sample is illuminated simultaneously from a subset of the angles. In this paper, we develop an algorithm for Fourier ptychographic microscopy with such multiplexed illumination. Specifically, we consider gradient descent type updates and propose an analytical step size that ensures the convergence of the iterates to a stationary point. Furthermore, we propose an accelerated version of our algorithm (with the same step size) which significantly improves the convergence speed. We demonstrate that the practical performance of our algorithm is identical to the case where the step size is manually tuned. Finally, we apply our parameter-free approach to real data and validate its applicability.

Introduction

Fourier Ptychographic Microscopy (FPM) is a powerful technique that overcomes the physical limits of a microscope objective by illuminating a sample from various angles using an LED array. This allows for the computational synthesis of "gigapixel" images with both a wide field-of-view and high spatial resolution. However, FPM faces challenges in speed and hyperparameter tuning. Sequential image capture is slow, and while multiplexed illumination speeds up data collection, it complicates the non-linear phase retrieval problem. Traditional reconstruction algorithms often require tedious manual tuning of the step size parameter to ensure convergence and speed.

Methods: Accelerated Wirtinger Flow

The authors address these issues by refining the Wirtinger Flow (WF) algorithm, a gradient descent variant for complex-valued optimization.

  • Analytical Step Size: A mathematical formula for the step size ($\mu$) is derived, showing it is inversely proportional to the maximum redundancy factor of the sampling in the Fourier domain. This makes the algorithm "auto-tuned" for any LED multiplexing pattern.
  • Nesterov Acceleration: An acceleration scheme inspired by Yurii Nesterov is adapted for the non-convex FPM problem, significantly boosting convergence speed.
  • Theoretical Guarantees: The paper provides a formal proof that the chosen step size ensures convergence to a stationary point.

Key Results

The algorithm was evaluated using physically accurate simulations and experimental data from a USAF resolution target captured with a Nikon TE300 microscope.

As illustrated in the convergence analysis above, the Accelerated Wirtinger Flow (AWF) reached the cost function floor significantly faster than standard WF. Notably, the performance of the analytical step size was identical to the manually optimized step size, validating the proposed formula.

On real experimental data, the algorithm improved spatial resolution by approximately 5$\times$. Multiplexed reconstruction using 200 images (a 30% reduction compared to sequential scans) yielded results very similar to the full sequential scan, demonstrating high data efficiency.

Key Insights

  • Redundancy as a Stability Metric: The stability of reconstruction is tied to Fourier-domain information overlap. Quantifying this overlap allows for a parameter-free approach.
  • Practical Acceleration: Nesterov acceleration provides massive practical speedups in this non-convex regime without compromising final image quality.
  • Accessibility: By removing manual tuning, the authors move FPM closer to being a practical tool for non-experts in optimization.

Limitations and Future Work

While the algorithm ensures convergence to a stationary point, it does not guarantee a global optimum. Future work could involve combining this accelerated approach with regularizers like Total Variation to mitigate artifacts in low-data regimes and pursuing real-time video-rate reconstruction for live-cell imaging.

Summary updated

Abstract

Fourier ptychographic microscopy enables gigapixel-scale imaging, with both\nlarge field-of-view and high resolution. Using a set of low-resolution images\nthat are recorded under varying illumination angles, the goal is to\ncomputationally reconstruct high-resolution phase and amplitude images. To\nincrease temporal resolution, one may use multiplexed measurements where the\nsample is illuminated simultaneously from a subset of the angles. In this\npaper, we develop an algorithm for Fourier ptychographic microscopy with such\nmultiplexed illumination. Specifically, we consider gradient descent type\nupdates and propose an analytical step size that ensures the convergence of the\niterates to a stationary point. Furthermore, we propose an accelerated version\nof our algorithm (with the same step size) which significantly improves the\nconvergence speed. We demonstrate that the practical performance of our\nalgorithm is identical to the case where the step size is manually tuned.\nFinally, we apply our parameter-free approach to real data and validate its\napplicability.\n
BibTeX
@inproceedings{bostan2018accelerated,
  author = {Emrah Bostan and Mahdi Soltanolkotabi and David Ren and Laura Waller},
  title = {{Accelerated Wirtinger Flow for Multiplexed Fourier Ptychographic\textbackslash\{\}n Microscopy}},
  booktitle = {ICIP},
  year = {2018},
  url = {https://openreview.net/forum?id=1NvVivRvfc9},
}
Figure