New AI chip identifies meibomian gland dysfunction within milliseconds

A research team from Peking Union Medical College Hospital and Tsinghua University has developed a compact optical AI chip that can identify meibomian gland dysfunction (MGD)-a major cause of dry eye disease-with more than 96% accuracy by analyzing the tissue's spectral signatures.

Reported in PhotoniX, the study demonstrates a new approach to ocular diagnostics by integrating a metasurface-based spectral convolutional neural network (SCNN) directly into an imaging sensor, allowing biochemical information to be captured in a single measurement.

A leading cause of dry eye, with limited diagnostics

MGD is the leading cause of evaporative dry eye, a condition affecting hundreds of millions of people worldwide. The disorder arises when lipid-secreting glands in the eyelids become obstructed or dysfunctional, destabilizing the tear film. Current diagnostic methods rely largely on clinical observation and functional tests, which can be subjective and may not detect early-stage disease. Imaging tools such as meibography provide structural information but do not capture molecular or compositional changes in glandular tissue.

Capturing spectral "fingerprints" in milliseconds

To address this gap, the researchers designed a miniaturized optical chip capable of extracting spectral features at the point of detection. The device integrates a dense array of metasurface filters onto a CMOS sensor. Each filter selectively modulates incoming light, enabling the system to perform optical computations-rather than digital post-processing-during image acquisition. This architecture allows spectral measurement and feature extraction to occur simultaneously, producing a full spectral feature map within tens of milliseconds. By contrast, conventional hyperspectral imaging systems typically require mechanical scanning and acquisition times on the order of seconds.

Distinct spectral signatures of diseased glands

The team applied the system to pathological tissue sections from individuals with and without MGD and identified wavelength-dependent differences between groups. In the visible range, MGD samples showed higher spectral responses, consistent with changes in hemoglobin-related optical properties that may reflect inflammation or altered microcirculation. In the near-infrared range, gland regions exhibited altered signals associated with lipid composition and tissue structure. These spectral features were also associated with clinical indicators of disease severity, suggesting potential relevance for objective assessment.

Improved accuracy over standard imaging

Using these spectral features for classification, the SCNN-based model achieved an average diagnostic accuracy of 96.22%, comparable to conventional hyperspectral imaging and substantially higher than models based on standard RGB images. Unlike RGB imaging, which captures only color and morphology, the spectral approach provides access to biochemical information, contributing to improved discrimination performance. In addition to accuracy, the chip offers a significant speed advantage, with acquisition times reduced by orders of magnitude compared to scanning-based systems.

Toward real-time ocular diagnostics

The chip is fabricated using CMOS-compatible processes and has a compact footprint, suggesting potential for integration into clinical imaging devices such as slit-lamp microscopes. With further development, the approach could enable rapid, objective, and quantitative assessment of meibomian gland function during routine eye examinations.