[Paper Review] Holographic Microscopy with Python and HoloPy

Solomon Barkley, Thomas G. Dimiduk+6 more|arXiv (Cornell University)|May 31, 2018

Digital Holography and Microscopy18 references3 citations

TL;DR

This paper presents HoloPy, a Python-based open-source framework that enables experimentalists to perform advanced 3D particle tracking in holographic microscopy using computational inference instead of traditional reconstruction. By wrapping legacy scattering codes and integrating with Python's scientific ecosystem, HoloPy allows fast, artifact-free inference of particle positions and shapes from holograms, significantly simplifying data analysis and accelerating discovery in soft matter and biological physics.

ABSTRACT

A holographic microscope captures interference patterns, or holograms, that encode three-dimensional (3D) information about the object being viewed. Computation is essential to extracting that 3D information. By wrapping low-level scattering codes and taking advantage of Python's data analysis ecosystem, HoloPy makes it easy for experimentalists to use modern, sophisticated inference methods to analyze holograms. The resulting data can be used to understand how small particles or microorganisms move and interact. The project illustrates how computational tools can enable experimental methods and new experiments.

Motivation & Objective

• To develop a user-friendly computational framework that enables experimentalists to extract 3D information from holograms without relying on artifact-prone reconstruction methods.
• To integrate legacy scattering simulation codes into a unified, extensible Python package for use in experimental data analysis.
• To accelerate scientific discovery by reducing the time and complexity required to analyze holographic time-series data from colloidal and biological systems.
• To support the analysis of complex, non-spherical, and clustered particles by enabling flexible scattering models within a consistent interface.
• To promote reproducibility and extensibility in holographic microscopy by providing a modular, well-tested, and open-source software stack.

Proposed method

• Uses a modular architecture to wrap low-level scattering codes (e.g., Mie theory, discrete dipole approximation) into a common Python interface.
• Employs iterative inference algorithms that fit simulated holograms to measured data by adjusting particle parameters (position, size, shape) to minimize residuals.
• Leverages Python’s scientific stack (NumPy, SciPy, scikit-image) for numerical computation, data processing, and visualization.
• Utilizes subprocess calls to interface with external command-line scattering codes, enabling integration of high-performance, legacy C/Fortran codes.
• Implements a unit testing framework to validate scattering model outputs against known analytical solutions (e.g., Lorenz-Mie for spheres).
• Supports parallelization via GPU or multi-core processing for individual hologram calculations, though sequential processing is preferred for time-series due to dependency on prior frames.

Experimental results

Research questions

• RQ1 Can a high-level Python package effectively abstract complex scattering physics and enable experimentalists to perform 3D inference without deep expertise in numerical methods?
• RQ2 How can legacy, high-performance scattering codes be efficiently and portably integrated into a modern, user-friendly scientific software stack?
• RQ3 To what extent does inference-based analysis outperform traditional reconstruction in terms of accuracy and artifact suppression for holographic microscopy?
• RQ4 How can computational efficiency be balanced with ease of use to support rapid experimental iteration in soft matter and biological systems?
• RQ5 Can a single scattering model be reused across diverse physical systems (e.g., colloids and rod-like bacteria) to enable cross-disciplinary discovery?

Key findings

• HoloPy successfully enables artifact-free 3D particle tracking by replacing traditional reconstruction with direct inference, eliminating common issues like blurring and fringe noise.
• The modular design allows seamless integration of diverse scattering models (e.g., spheres, ellipsoids, clusters), enabling analysis of non-spherical and complexly shaped particles.
• The framework supports rapid prototyping and analysis of holographic time-series, with results typically available within a day, allowing for iterative experimental design.
• Validation against analytical solutions (e.g., Lorenz-Mie theory) confirms the accuracy of the wrapped scattering models across various particle geometries and sizes.
• The use of inference over reconstruction has enabled new scientific discoveries, such as previously unobserved aspects of bacterial motility in rod-shaped bacteria.
• The conda-forge distribution and automated testing framework ensure cross-platform compatibility and long-term maintainability, supporting broad adoption in experimental labs.