$\textit{In vivo}$ fundus imaging and computational refocusing with a diffuser-based fundus camera

Corey Simmerer, Marisa M. Morakis, Lia Gomez-Perez, T. Y. Alvin Liu, Nicholas J. Durr, Durr, Nicholas J.arXiv 2024 · DOI · arXiv

Abstract

Significance: Access to diagnostic eye care could be expanded with high-throughput and easy-to-use tools. Phase mask-based imaging may improve the fundus camera by enabling computational refocusing with no moving parts. While phase mask-based imaging has been demonstrated in a model eye, this approach has not been shown $\textit{in vivo}$. Aim: A computational fundus camera was designed, constructed, and evaluated with the goal of determining the feasibility and performance of phase mask-based computational imaging of the $\textit{in vivo}$ fundus. Approach: A holographic diffuser was introduced in a modified commercial fundus camera at a plane conjugate to the ocular pupil, resulting in a linear and shift-invariant point spread function that varies with refractive error. The image could be digitally refocused across a range of $\geq\pm$ 10 diopters of defocus error. The device was tested for ocular safety, and a human imaging pilot study was performed. Results: The device captured and digitally refocused color human fundus images. The field of view was $\geq$35 degrees and resolution was 7.7-9.6 line pairs per mm. Conclusions: We present the first $\textit{in vivo}$ diffuser-based fundus images, demonstrating the feasibility of computational imaging for ocular diagnostics.

Introduction

Imaging the fundus—the interior surface of the eye including the retina and optic nerve—is essential for diagnosing conditions like glaucoma and diabetic retinopathy. Conventional fundus cameras, however, require precise manual focusing to compensate for a patient's refractive error (myopia or hyperopia). This mechanical adjustment necessitates skilled operators and complicates the design of portable screening tools.

Computational imaging offers a solution by replacing mechanical focusing with a phase mask and digital reconstruction. While previously demonstrated in laboratory model eyes, this study presents the first successful $\textit{in vivo}$ application of a diffuser-based fundus camera capable of massive digital refocusing.

Methods and System Design

The researchers modified a commercial Canon CR-DGi fundus camera by integrating a custom "Diffuser Camera" module. The core of this system is a holographic diffuser placed at a plane conjugate to the ocular pupil.

Overview of the computational fundus imaging pipeline, including imaging, calibration, deconvolution, and focal stack reconstruction.

Optical Principles

By placing the diffuser at the pupil conjugate, the system maintains a linear and shift-invariant (LSI) property. When the eye is out of focus, the resulting point spread function (PSF) on the sensor scales in size rather than simply blurring. This predictable scaling allows the system to capture information about the fundus across a wide range of focal depths in a single exposure.

Hardware Implementation

The system utilizes sequential red, green, and blue LED illumination to capture color information on a monochromatic sensor. The optical path was carefully designed to bypass corneal reflections, which are significantly brighter than the retinal signal, by using the commercial camera's existing annular illumination and holed mirror system.

Hardware implementation and optical path diagram of the modified Canon CR-DGi fundus camera.

Computational Reconstruction

Because the patient's exact refractive error is typically unknown at the time of capture, the system generates a "focal stack." It applies Tikhonov regularization in the Fourier domain using a library of pre-calibrated PSFs corresponding to different dioptric powers. The sharpest image is then selected from this stack.

Results

The prototype was validated through both model eye experiments and a pilot human study.

  • Refocusing Capability: The system demonstrated the ability to digitally refocus images across a range of $\geq \pm 10$ diopters. This range is sufficient to cover the vast majority of human refractive errors without any mechanical movement of lenses.
  • Resolution and Field of View: The measured resolution was 7.7–9.6 line pairs per mm, and the field of view (FOV) reached $\geq 35$ degrees. While the resolution is lower than that of high-end clinical cameras, it is sufficient for identifying key anatomical features like the optic disk and major blood vessels.
  • Safety: The device was confirmed to meet ISO 15004-2:2007 light safety standards for Group 1 (non-hazardous) ophthalmic instruments.

Discussion and Conclusion

This work represents a significant milestone in computational ophthalmology by proving that diffuser-based imaging is viable for $\textit{in vivo}$ human fundus photography. The ability to "deskill" the focusing process could lead to more accessible, automated screening devices for global health.

Future work will focus on closing the resolution gap compared to conventional cameras. Potential improvements include optimizing the phase mask design (e.g., using microlens arrays) and developing more sophisticated reconstruction algorithms for dense scenes. Additionally, integrating a fixation target would improve patient alignment and help control instrument-induced myopia.

Ultimately, this technology paves the way for a new generation of "point-and-shoot" fundus cameras that leverage computational power to simplify ocular diagnostics.

Summary updated

Abstract

Significance: Access to diagnostic eye care could be expanded with high-throughput and easy-to-use tools. Phase mask-based imaging may improve the fundus camera by enabling computational refocusing with no moving parts. While phase mask-based imaging has been demonstrated in a model eye, this approach has not been shown $\textit{in vivo}$. Aim: A computational fundus camera was designed, constructed, and evaluated with the goal of determining the feasibility and performance of phase mask-based computational imaging of the $\textit{in vivo}$ fundus. Approach: A holographic diffuser was introduced in a modified commercial fundus camera at a plane conjugate to the ocular pupil, resulting in a linear and shift-invariant point spread function that varies with refractive error. The image could be digitally refocused across a range of $\geq\pm$ 10 diopters of defocus error. The device was tested for ocular safety, and a human imaging pilot study was performed. Results: The device captured and digitally refocused color human fundus images. The field of view was $\geq$35 degrees and resolution was 7.7-9.6 line pairs per mm. Conclusions: We present the first $\textit{in vivo}$ diffuser-based fundus images, demonstrating the feasibility of computational imaging for ocular diagnostics.
BibTeX
@misc{simmerer2024textit,
  author = {Corey Simmerer and Marisa M. Morakis and Lia Gomez-Perez and T. Y. Alvin Liu and Nicholas J. Durr and Durr, Nicholas J.},
  title = {{$\textbackslash\{\}textit\{In vivo\}$ fundus imaging and computational refocusing with a diffuser-based fundus camera}},
  year = {2024},
  eprint = {2406.00122},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2406.00122},
}

Loading related content...

Figure