What Is Fundus Photography? A Camera Module Guide for OEM Retinal Imaging Systems
Updated: Nov 12
Fundus photography is the capture of the back of the eye through the pupil, producing a single image a clinician reads for disease. The eye is the only place in the body where blood vessels can be seen directly, which is why that one photograph can reveal diabetes, hypertension, and glaucoma.
Taking it is harder than it sounds. You shoot through a hole a few millimeters wide, light the subject through the same hole, and keep the light low enough that the patient can tolerate it. This post answers what fundus photography is from an engineering point of view, then works through the camera specifications that decide whether the image is diagnostic or merely pretty.
What Is Fundus Photography, and What Does the Image Have to Show What Is Fundus Photography and What the Image Has to Show
Four things have to be visible. The optic disc, read for cupping and pallor. The macula, the small central region responsible for detailed vision. The retinal blood vessels, read for calibre, tortuosity and haemorrhage. And the surface between them, where lesions appear.
Color fundus photography is the standard form, and color is not decorative: haemorrhages, exudates, and drusen are distinguished partly by hue, so a camera that shifts reds or crushes greens removes information the grader needs.
diabetic retinopathy screening is the largest use by volume, with national programmes photographing millions of eyes a year. Glaucoma, macular degeneration, hypertensive retinopathy and papilloedema all rely on the same image.
Why the Fundus Is a Hard Imaging Problem
This is the part most articles skip, and the reason a general-purpose camera cannot simply be pointed at an eye.
You illuminate and image through the same pupil, an aperture that may be only 3 mm wide undilated. Every surface on the way in- the cornea, the crystalline lens, the objective itself- reflects some of that light straight back into your imaging path, and those reflections are brighter than the retinal image you want. A fundus imaging system separates the two paths geometrically, usually with an annular beam entering through the outer ring of the pupil while the image exits through the centre. Get the geometry wrong, and you photograph a bright crescent instead of a retina.
The second constraint is light level. You are shining light into a living eye, so illumination is limited on comfort and safety grounds, and the retina is not very reflective. A low-light medical camera is therefore the baseline requirement, not a nice-to-have.
The third is that you get one attempt. The pupil constricts after a flash, the patient blinks, and in a non-mydriatic fundus camera you cannot simply fire again straight away. That is why a non-mydriatic fundus camera leans so hard on sensitivity: the capture has to be right the first time, at whatever light level the eye will tolerate.
Where the Camera Module Sits, and Where It Does Not
The optical chain runs: eye, objective optics, path separation, relay optics, camera module, host.
A fundus camera module occupies one box in that chain. It receives the image the optics have already formed and converts it to data. It does not form the retinal image, control the illumination geometry, or manage pupil alignment.
That boundary matters enough to state plainly. These modules are the imaging front end of an OEM fundus system. The retinal optics, illumination path, pupil alignment, and eye-safety engineering remain the device maker’s responsibility, and a module specification alone cannot establish that a finished instrument is safe or diagnostic.
Within it, the module still decides a great deal: how much light becomes signal rather than noise, whether colours land where the grader expects, how much of the field resolves, and how the image reaches the host. Our guide on how a fundus camera works covers the optics and illumination side.
So the honest answer to what camera is used in fundus photography is that there is no fundus sensor. There are sensors whose format, sensitivity, and color behavior suit the job, behind optics that do the hard part.
Sensor Specifications That Decide Fundus Image Quality
Megapixels are the number on the box and the least useful one. Four other properties matter more.
Optical format and pixel size come first: A larger format collects more light and pairs more easily with relay optics. Between two 8MP sensors, the larger format almost always gives the better fundus image, because each pixel gathers more photons in a scene short of them.
Sensitivity, including NIR: Back-illuminated pixels put the wiring behind the photodiode, clearing the light path. nir sensitivity matters separately, because many instruments use near-infrared illumination for alignment: the pupil does not constrict to light it cannot see. Poor NIR response makes the alignment preview useless even when the colour capture is fine.
Color fidelity: Grading depends on hue, so a medical imaging camera module needs a predictable colour response rather than a pleasing one. RAW output is worth having, since it lets your pipeline apply a transform tuned for retinal tissue instead of for skin and sky.
Dynamic range: The optic disc is bright, and the peripheral retina is dark, and both must be readable in one frame. HDR helps, though a single-flash capture mostly relies on native dynamic range.
MTF and signal-to-noise sit behind all four. A 20MP sensor behind soft optics resolves less real detail than a 13MP sensor behind good ones.
Field of View and Resolution: What 45 Degrees Actually Demands
Here is the arithmetic that turns “how many megapixels” from an opinion into a calculation.
Retinal magnification is roughly 290 micrometres of retina per degree of visual angle. A fundus camera field of view of 45 degrees, the screening standard, therefore covers about 13 mm of retina.
Divide by your sensor. A 13mp camera module such as the AR1335 gives 4208 pixels across its long axis, or about 93 pixels per degree. A 20mp camera module such as the AR2020 gives roughly 5100 pixels, about 113 pixels per degree.
Apply that to a lesion. A 100 µm microaneurysm subtends 0.34 degrees and lands on roughly 32 pixels at 13MP, 39 at 20MP. The smallest microaneurysms, around 25 µm, subtend 0.086 degrees and land on about 8 pixels at 13MP, rising to 10 at 20MP.
That is the real argument for resolution, and it is narrow. Eight pixels is marginal for a human grader and thin for a classifier; ten is better; neither is transformative. A retinal imaging camera at 20MP is worth specifying for detailed work, but it is not a substitute for good optics.
Widen the field, and it tightens further. The same 13MP sensor across 60 degrees drops to about 70 pixels per degree, and the 25 µm lesion falls to 6 pixels. Wide-field imaging buys coverage and sacrifices resolution, every time.
Focus: Autofocus or Fixed, and Who Does the Focusing
In a fundus camera, what moves the image is not subject distance but the patient’s refraction. A myopic eye and a hyperopic eye place the retinal image in different planes, and the instrument compensates across roughly fifteen diopters either side of neutral. Two architectures do that, and both are legitimate.
The optics compensate. A focusing group inside the instrument shifts the image plane, and the module stays in focus, sensor locked to that plane. This is how most desktop clinical instruments work, and an autofocus camera module is unhelpful here because its hunting fights the operator.
The module compensates. Its own lens focuses, driven by contrast detection on the retinal image. Common in handheld fundus camera and portable fundus camera designs, where the operator is a technician rather than an ophthalmic photographer.
On that route, VCM autofocus is what you specify: a voice coil moves the lens against a spring, and the loop settles on the sharpest position. What matters is not that it has autofocus but that you can command an absolute lens position and repeat it, so every patient gets the same focus.
One caution. Contrast-detection autofocus tuned for general scenes will lock onto a corneal reflection or the pupil edge instead of the retina, so the focus window has to be constrained to the retinal image. That is integration work on your side.
Interface: USB 3.2 Gen 1 or MIPI CSI-2
A fundus camera’s data profile makes this easier than it looks. You run a low-resolution preview while the operator aligns on the pupil, then capture one full-resolution frame at the flash. You need bandwidth for a large frame in a burst, not a sustained frame rate.
MIPI CSI-2 suits an integrated design: With the processor on the same board stack, you get the lowest latency and power and the cheapest bill of materials, at the cost of a cable measured in centimetres and driver work against your SoC. A MIPI camera for medical device work suits a handheld unit built around an embedded processor.
USB 3.2 Gen 1 suits a host-attached design: Five gigabits per second carries a 20MP still comfortably, and UVC means Windows and Linux see the camera with no driver. A USB camera for fundus imaging is the lower-risk path when the host is a PC or an SBC.
Rule out USB 2.0 explicitly. At 480 Mbps, it forces MJPEG on the capture path, and JPEG artefacts land precisely on the vessel edges a grader is reading.
Vadzo Camera Modules for Fundus Integration
Every module below is a colour autofocus ophthalmic camera module candidate rather than a finished instrument. The boundary stated earlier applies to all of them: the optics, illumination, and eye-safety engineering are yours.
Falcon Series: USB Modules
Module | Sensor | Resolution | Focus | Buy Now |
Falcon-1335CRA | onsemi AR1335 | 13MP | VCM autofocus, 100 mm to infinity | |
Falcon-1335CRO | onsemi AR1335 | 13MP | VCM autofocus with OIS | |
Falcon-258CRA | Sony IMX258 | 13MP | PDAF and contrast autofocus | |
Falcon-821CRH | onsemi AR0821 | 8MP, 4K HDR | VCM autofocus | |
Falcon-830CRH | onsemi AR0830 Hyperlux LP | 8MP, 4K HDR | VCM autofocus |
Bolt Series: MIPI CSI-2 Modules
Reading the tables. The AR2020 rows are the resolution option, and that sensor’s larger format and enhanced NIR response make it the strongest starting point for a detailed fundus camera design. The AR1335 rows are the mainstream 13MP configuration, most commonly specified for color fundus photography. The AR0821 and AR0830 rows trade resolution for HDR, which suits scenes where disc-to-periphery contrast limits you rather than lesion size.
The OIS variants get a caveat rather than a recommendation. Stabilisation earns its place in a handheld instrument; in a chin-rest device the head is already fixed, and OIS adds a moving element you do not need.
An embedded camera for an ophthalmology programme normally tests two or three of these against the same optics rather than choosing from a datasheet. Treat every ophthalmic camera module here as a candidate to evaluate, not a specification to trust.
Validation Checklist Before You Commit
Run these against an evaluation unit behind your own optics, not against a datasheet.
Confirm your relay optics cover the sensor without corner vignetting, and check sharpness across the field rather than only at centre. Measure colour response against a calibrated target under your own illumination, since your LEDs will dominate the result. Check NIR response at the wavelength your alignment path uses. Capture at the exposure and gain your illumination budget allows, then look at the noise floor in the dark periphery, where it shows first.
Constrain the autofocus window to the retinal image and verify it does not lock onto corneal reflections. Confirm you can command and repeat an absolute lens position. Then put the captures in front of clinicians and ask whether they would grade from them, which is the only test that counts.
Request an evaluation unit and run it on your own optical bench before committing. Our engineering team can shortlist against your field of view, illumination budget, and host architecture.
Frequently Asked Questions
What is fundus photography?
Fundus photography is the capture of the back of the eye through the pupil, producing an image a clinician reads for disease. The optic disc, the macula, and the retinal blood vessels all have to be visible and correctly rendered, since grading depends on their shape, caliber, and color. It is the backbone of diabetic retinopathy screening programmes, where color fundus photography in a non-mydriatic fundus camera is the standard capture, and it is also used for glaucoma, macular degeneration, and hypertensive retinopathy. The optics do the hard part, and the camera module converts the image that those optics form into data. Vadzo supplies that module as a board-level fundus camera module for OEM instruments rather than as a finished camera, so an embedded camera for an ophthalmology programme starts from a qualified ophthalmic camera module and builds the optics around it.
How does a fundus camera work?
A fundus camera lights and images the retina through the same pupil at the same time: illumination enters through the outer ring of the pupil while the retinal image leaves through the centre, which keeps corneal and lens reflections out of the imaging path. That separation is the whole of how a fundus camera works. Objective and relay optics then form the image on the sensor of a retinal imaging camera, and a host processes and stores it. Light levels stay deliberately low for patient comfort and eye safety, which is why sensitivity dominates the sensor choice. Vadzo’s Falcon and Bolt modules sit at the end of that chain as the imaging front end of a fundus imaging system.
What camera is used in fundus photography?
There is no fundus-specific sensor, so what camera is used in fundus photography comes down to format, sensitivity, and color behavior behind the right optics. A medical imaging camera module for this work wants a large optical format, back-illuminated pixels, good nir sensitivity for the alignment path, and predictable colour response, because a low-light medical camera that shifts hue removes grading information. Resolution matters second: a 13mp camera module gives about 93 pixels per degree across a 45-degree field, and a 20mp camera module about 113. The interface is the last decision, between a USB camera for fundus imaging and a MIPI camera for medical device integration. Vadzo offers the same sensors on both interfaces, so the two choices stay independent.
How many megapixels does a fundus camera need?
Between 13MP and 20MP, and you should work it from the lesion rather than the marketing number. At the 45 degree fundus camera field of view used for screening, 13MP resolves a 100 µm microaneurysm across roughly 32 pixels and a 25 µm one across roughly 8, while 20MP raises those to about 39 and 10. Widening to 60 degrees drops the smaller lesion to about 6 pixels, so fundus camera field of view and resolution have to be specified together. Beyond that, optical quality limits what any sensor can deliver, and an AI fundus camera pipeline is more sensitive to noise and colour consistency than to raw pixel count, which is why an AI fundus camera rarely benefits from the jump to 20MP as much as the datasheet suggests. Vadzo’s AR2020 modules are the 20MP option and the AR1335 modules the 13MP one, on either interface.
Does the camera module need autofocus?
Only if the module’s lens is what compensates for the patient’s refraction. If a focusing group inside the optics does that instead, specify fixed focus and lock the sensor plane, because an autofocus camera module will fight the operator’s adjustment. If the module’s own lens is the focusing mechanism, which is usual in a handheld fundus camera or a portable fundus camera built for tele-ophthalmology work, where tele-ophthalmology programmes put the instrument in the hands of a technician rather than a photographer, then VCM autofocus with absolute, repeatable lens positioning is what you need. Either way, the focus window must be constrained to the retinal image so it does not lock onto a corneal reflection. Every Vadzo module listed here is an autofocus variant, and our engineering team can advise on fixed-focus configurations for optics-compensated designs.




