Industry News
09 Nov 2026

Machine Learning and Long-Range OCT Map Layer-Specific Lens Deformation During Accommodation

Machine Learning and Long-Range OCT Map Layer-Specific Lens Deformation During AccommodationA study using long-range OCT and machine learning has mapped the human lens layer-by-layer during accommodation, revealing that deformation is highly localised to the nucleus and inner cortex.

When your patients shift their gaze from a distant street sign to the text on their smartphones, the crystalline lens inside each eye reshapes itself in a fraction of a second. For decades, clinicians and researchers have described this remarkable feat using global metrics, measuring total lens thickness and front-surface curvature. 

But a study published in Eye and Vision argues that this traditional, whole-lens view misses where the real action happens. 

Using a combination of long-range optical coherence tomography (OCT) and machine learning–assisted image segmentation, researchers have successfully mapped accommodation layer-by-layer through the full thickness of the human lens. The findings challenge the assumption of the lens as a uniform structure, proving that accommodative deformation is heavily concentrated in the lens nucleus and the innermost cortical layers, while the outer cortex barely moves.

Beyond the 'Homogeneous Blob'

The crystalline lens is a highly stratified organ. It features a dense central nucleus surrounded by a cortex, with cross-sectional imaging revealing alternating light and dark bands known as optical signal discontinuity (OSD) zones. Under the Oxford clinical cataract classification, the cortex is divided into four zones on each side of the nucleus (C1 through C4, with C4 lying innermost). 

Historically, segmenting these subtle bands automatically has been a persistent obstacle, limiting in vivo studies to global biometric measurements. To overcome this, the research team—led by investigators from Nicolaus Copernicus University and Gdansk University of Technology—built a sophisticated deep learning pipeline.

Using a U-Net architecture equipped with a pre-trained ResNet-50 encoder, the model achieved a mean intersection over union (IoU) of 0.936 on an unseen test set, with exceptional recall (0.974) and specificity (0.995). The predicted surfaces were then fed into a custom ray-tracing algorithm that corrected the OCT images for light refraction across the cornea, aqueous, cortex, nucleus, and vitreous.

The Methodology: Capturing the Full Eye

The study enrolled 60 healthy young adults (mean age 26, range 18–50). Each participant had one eye imaged using a commercial long-range swept-source OCT system operating at 60,000 A-scans per second. 

Crucially, the system boasted an axial imaging range of 40.2 millimetres and an axial resolution of 11 micrometres, long and sharp enough to capture the entire ocular axis in a single acquisition. Participants viewed targets through a modified Badal optometer, allowing the team to set accommodative demands of 0, 2, 4, and 6 dioptres without altering the apparent size of the target.

The Breakthrough: Localised Deformation

While the global biometric results matched decades of prior literature, anterior chamber and vitreous chamber depths shortened, axial length increased slightly (interpreted as transient mechanical deformation), and the pupil constricted steadily by roughly 81.66 micrometres per dioptre, the lens itself told a much sharper, more complex story.

Total lens thickness increased linearly at about 24 µm per dioptre, but the nuclear thickness grew significantly faster at 36 µm per dioptre. Meanwhile, the combined cortical thickness shrank at a rate of -14 µm per dioptre. 

The zonal analysis located this deformation with unprecedented precision:

  • The Outer Cortex (C1–C3): Remained remarkably stable. Across most of these peripheral zones, mean thickness changes were 5 µm or less, representing less than a 3% change even at the maximum 6-dioptre demand.
  • The Inner Cortex (C4): Acted in direct coordination with the nucleus. Both the anterior and posterior C4 zones thinned monotonically and significantly. At 6 dioptres, the inner cortical layers thinned by a staggering 22% to 25%.
  • Curvature: The anterior lens and nuclear radii steepened significantly (decreasing by ~289 µm/D and ~123 µm/D, respectively), while the corneal radii remained essentially flat, confirming the cornea is largely decoupled from accommodative effort in healthy eyes.

In short, the lens does not swell uniformly when it focuses. The dense central mass expands and pushes outward, squeezing the inner cortical shell thinner, while the outer cortex acts as a mechanically supportive boundary.

Novel Sex Differences in Accommodation

The study's massive dataset also uncovered subtle, previously unreported sexual dimorphisms in the accommodating eye. 

While males naturally exhibited longer vitreous chambers and greater axial lengths as stable anatomical offsets, females displayed a thicker anterior C4 zone, slightly smaller nuclear radii, and a modestly higher rate of axial elongation per dioptre of accommodation. The authors note these subtle structural differences warrant further investigation, particularly in the context of individualised biomechanical modelling.

What This Means for Eyecare Professionals

For optometrists, ophthalmologists, and optical engineers, the clinical and industry implications are profound. 

Firstly, it confirms that accommodation cannot be fully described by global lens metrics alone. Biomechanical and optical models of the eye, which are critical for the next generation of accommodating intraocular lenses (IOLs) and presbyopia treatments, must incorporate cortical and nuclear behaviour rather than treating the lens as a single elastic lump.

Secondly, the OSD-based zonal analysis offers a novel, depth-resolved view of the lens, indicating that these pre-existing discontinuity architectures act as a layered deformation framework. This establishes a vital quantitative baseline for future studies investigating age-related changes, presbyopia, and accommodative dysfunction.

Limitations to Note

The researchers are careful to caveat their findings. The cohort consisted of healthy young adults with low refractive error, meaning the results may not directly translate to highly myopic, hyperopic, or astigmatic eyes, where accommodative dynamics can differ significantly. 

Additionally, the curvature values are best-fit spheres over the central 3mm zone, which may miss complex aspheric shape changes at higher demands. The study also utilised theoretical phase refractive indices, which the authors calculate overestimates absolute thicknesses by roughly 0.65% to 0.72%, a uniform scaling that does not affect the relative structural deformations reported, but highlights the need for future studies to adopt newly established empirical group refractive indices.

Nevertheless, the central message is clear: the human lens is a marvel of layered engineering. By combining long-range OCT with advanced AI, researchers have finally peeled back the curtain on how our eyes truly focus up close, providing the eyecare industry with a highly detailed, layered target for the innovations of tomorrow.