📄 bioengineering

Quantitative evaluation of LED based optical autofocus module

This paper presents an improved, open-source LED-based optical autofocus module for light microscopy and a 2D autocorrelation method to quantify its performance, demonstrating that the new system achieves axial stability with a standard deviation of less than 10 nm over 45 minutes despite initial power variations.

Original authors: Habte, S., Kumar, S., Lightley, J., Garcia, E., Neil, M., French, P. M.

Published 2026-04-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Habte, S., Kumar, S., Lightley, J., Garcia, E., Neil, M., French, P. M.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine you are trying to take a perfectly sharp photo of a tiny, delicate flower using a powerful microscope. But there's a problem: the microscope is like a wobbly table. As time passes, the room heats up, the building settles, or the machine itself warms up, causing the flower to drift slightly out of focus. If you don't fix this, your photo (or your scientific data) becomes blurry and useless.

This paper is about building a smart, self-correcting camera system that keeps the flower in focus automatically, even when the machine is shaking or drifting. The researchers didn't just build it; they built a way to prove it works better than the old ways, and they did it using cheap, safe parts instead of expensive, dangerous ones.

Here is the story of their discovery, broken down into simple concepts:

1. The Problem: The "Drifting" Microscope

In high-tech microscopy, scientists need to keep the sample perfectly still in the "focal plane" (the sweet spot where everything is sharp).

The Old Way: Some microscopes take a quick "test photo" of the sample to see if it's blurry, then move the lens. But this is slow and can damage the sample with too much light.
The "Optical Autofocus" (OAF) Way: Instead of looking at the sample, these systems shoot a separate, invisible beam of light at the glass slide. If the slide moves, the reflection changes. The computer sees this change and moves the lens back into place instantly.

2. The Upgrade: Swapping the Laser for a Flashlight

The researchers had a previous version of this system that used a Super Luminescent Diode (SLD). Think of this like a very expensive, fragile, high-tech laser pointer. It worked well, but it was costly and required strict safety rules (like wearing special goggles) because it was a laser.

The Innovation: They replaced the expensive laser with a simple LED (like the light in a cheap flashlight or a phone screen) and a piece of multimode fiber optic cable (a bundle of tiny glass strands).

Why? LEDs are cheap, robust, and safe. You don't need special safety gear to use them.
The Catch: Because the LED light is "messier" (less organized) than the laser, it behaves differently when it gets hot.

3. The "Warm-Up" Surprise

When they turned on their new LED system from a cold start, something weird happened.

The Analogy: Imagine you turn on a car engine. For the first 10 minutes, the engine is cold, the oil is thick, and the car idles roughly. As it warms up, it runs smoothly.
What happened: As their LED warmed up, its light output dropped slightly (by about 2%). The autofocus computer didn't know this was happening. It thought the change in light meant the sample had moved, so it kept adjusting the lens unnecessarily. This caused the microscope to drift out of focus by over half a millimeter—a huge error in the microscopic world!

4. The Solution: The "Smart Dimmer Switch"

The researchers realized the computer was getting confused by the changing brightness of the LED.

The Fix: They taught the software to constantly check the brightness of the light source. If the LED gets dimmer as it warms up, the software says, "Ah, the light is just getting weaker, not the sample moving!" and it normalizes (adjusts) the calculation.
The Result: Once they added this "smart dimmer," the system became incredibly stable. It could hold the focus with a precision of less than 10 nanometers (that's thinner than a human hair by a factor of 10,000) for over 45 minutes, even while the LED was warming up.

5. The "Magic Mirror" Test (How they proved it worked)

How do you know a self-driving car is actually driving well if you can't see the road? You need an independent test.

The Old Test: They used to take stacks of photos of tiny beads to see how blurry they were. This was slow and tedious.
The New Test (The "Magic Mirror"): They used a special trick called astigmatic imaging. Imagine looking at a star through a slightly warped window. The star looks like a circle when in focus, but turns into a horizontal line or a vertical line when out of focus.
- They took a picture of a glowing bead.
- They used a computer algorithm (like a digital magnifying glass) to measure the shape of that bead's "glow."
- If the glow was a perfect circle, they were in focus. If it was an oval, they knew exactly how much to move the lens.
- This was fast, easy, and gave them a "gold standard" to prove their new LED system was working perfectly.

6. The "Infinity Alignment Tool"

Finally, they built a simple tool to help other scientists set up their microscopes correctly.

The Analogy: It's like a "laser level" for photographers. It ensures that the camera and the lens are perfectly aligned so that when you look through the microscope, the image is sharp from edge to edge. This tool uses a simple grid pattern to make sure everything is lined up "at infinity" (perfectly parallel), saving hours of fiddling.

The Big Picture

This paper is a victory for open-source science.

Cheaper: They replaced a $1,000+ laser with a $50 LED.
Safer: No more laser safety goggles required.
Smarter: They fixed a hidden bug (the warm-up drift) that would have ruined experiments.
Accessible: They made all their code and designs free for anyone to download, helping scientists in labs with fewer resources build high-end microscopes.

In short, they built a "self-driving" microscope that is cheaper, safer, and more reliable than before, and they invented a clever way to prove it's working without breaking a sweat.

1. Problem Statement

Optical microscopy, particularly for high-throughput applications (e.g., whole-slide scanning, high-content analysis) and super-resolution techniques (e.g., Single Molecule Localisation Microscopy - SMLM), requires robust autofocus (AF) systems to counteract mechanical and thermal drift.

Limitations of Existing Solutions:
- Software-based AF: Relies on image quality metrics (sharpness, entropy) which require acquiring a reference z-stack for every field of view, significantly slowing acquisition. They often fail with thick or evolving samples.
- Optical AF (OAF) Hardware: While faster, commercial or custom OAF modules often rely on expensive, fragile, or laser-safety-regulated light sources like Super Luminescent Diodes (SLDs) or laser diodes.
- Performance Quantification: There is a lack of rapid, independent methods to quantify AF performance. Previous methods (e.g., acquiring bead z-stacks or using test charts) are time-consuming or do not accurately replicate biological sample conditions (e.g., low reflectivity at coverslip interfaces).
- Thermal Drift: Light sources (especially LEDs) often exhibit power fluctuations as they reach thermal equilibrium, which can introduce systematic errors in AF calculations if not accounted for.

2. Methodology

The authors developed and evaluated an improved, open-source optical autofocus module ("openAF") and a novel method for independent performance quantification.

A. Hardware Modifications (LED-based openAF)

Light Source Replacement: The previous SLD/Single Mode Optical Fiber (SMOF) source was replaced with a low-cost 850 nm LED coupled into a 50 µm core Multimode Optical Fiber (MMOF).
Optical Path: The LED/MMOF beam is collimated using two orthogonal cylindrical lenses with different focal lengths. One lens is slightly translated to create an astigmatic focus shift, allowing the determination of the defocus sign from a single image.
Noise Reduction: A 3D-printed rectangular aperture was added to block stray light from the MMOF cladding.
Algorithm Improvement: To address power variations, the authors implemented a continuous background intensity normalization. The algorithm scales the pre-recorded background image based on the real-time average intensity of the autofocus camera before subtraction, correcting for LED power drift.

B. Independent Performance Quantification (Astigmatic Autocorrelation)

To measure AF accuracy independently of the AF system itself, the authors utilized:

Astigmatic Imaging: A cylindrical lens was inserted into the fluorescence imaging path to induce astigmatism in the Point Spread Function (PSF) of 100 nm fluorescent nanobeads.
2D Autocorrelation Analysis: Instead of localizing individual beads (which is computationally heavy), they calculated the 2D autocorrelation function of the bead image across the Field of View (FOV).
Defocus Metric: The ratio of the widths ( $\sigma_X / \sigma_Y$ ) of Gaussian fits to the orthogonal mean intensity projections of the central autocorrelation peak provides a monotonic relationship with defocus. This allows for rapid, single-image defocus estimation within a range of $\pm 0.7$ µm.

C. Alignment Tool

A custom "Infinity Alignment Tool" was developed to rapidly align the imaging arm (tube lens to camera) and the autofocus arm, ensuring the graticule is in the focal plane of the objective and the camera is at the "circle of least confusion" for astigmatic systems.

3. Key Contributions

Cost-Effective LED Implementation: Successfully replaced the expensive SLD with an LED/MMOF configuration, removing laser safety concerns and reducing costs while maintaining robust performance.
Novel Quantification Method: Introduced a rapid, software-based method using astigmatic autocorrelation to independently measure AF drift with high precision, avoiding the need for time-consuming z-stack acquisitions.
Power Normalization Algorithm: Identified and corrected a systematic error caused by LED power drift during thermal warm-up. By normalizing the background subtraction against real-time intensity, they achieved high stability immediately after switching on the system.
Open-Source Ecosystem: Provided open-source hardware designs, software (openAF plugin for µManager), and analysis tools, promoting accessibility for custom microscope builders.

4. Results

Stability Performance:
- Without Normalization: The LED-based system showed a focus drift of >500 nm over 20 minutes as the LED warmed up and power dropped by ~2%.
- With Normalization: The modified algorithm achieved an axial stability standard deviation (SD) of <10 nm (specifically ~7 nm) over 45 minutes, even while the LED power varied by ~2%.
- SLD Comparison: The previous SLD-based system showed a drift of >100 nm without normalization (due to ~5% power variation) but improved to ~8 nm SD with normalization.
Operating Range: The LED/MMOF system has a closed-loop operating range of $\pm 15$ µm, which is sufficient for time-lapse imaging at a single field of view. The range is asymmetric due to optical offsets but covers the depth of field for a 100x 1.4 NA objective.
Accuracy: The defocus prediction error was significantly less than the Depth of Field (DOF) of the objective lens (200 nm), confirming the system's ability to maintain focus within the diffraction limit.

5. Significance

Accessibility: The shift to LED sources makes high-performance optical autofocus accessible to researchers in lower-resourced settings and eliminates the regulatory burden of laser safety.
Robustness for Biological Imaging: The system performs reliably even with high-NA oil immersion objectives where the coverslip/aqueous interface reflectivity is extremely low (~0.4%), a common challenge for OAF systems.
Enabling Advanced Microscopy: The demonstrated stability (<10 nm) makes this module suitable for demanding applications like 3D SMLM and long-term time-lapse assays where nanometer-scale drift correction is critical.
Methodological Advance: The autocorrelation-based quantification tool provides a standard, rapid benchmark for evaluating any optical autofocus system, independent of the specific hardware implementation.

In conclusion, this work presents a practical, low-cost, and highly stable solution for optical autofocus, validated by a rigorous, independent quantification method that addresses the critical issue of thermal drift in light sources.