The Latent
Neural Space.
How Clouseau transforms raw pixels into logic.
Powered by state-of-the-art transformers.
14x14 token patches
Chapter 01
From Pixels to Patches.
Clouseau doesn't look at an image as a whole; it slices the retinal scan into 14x14 patches. Each patch is then flattened into a single, high-dimensional vector.
These vectors are the atoms of clinical intelligence, completely removing clinical noise so the models can lock onto the core biological indicators required for critical triaging.
Chapter 02
The Eight Streams of Clinical Logic.
Neural logic isn't linear. Clouseau uses Multi-Head Attention to analyze the retinal space through eight parallel lenses simultaneously.
One "head" focuses on vascular integrity, another on the optic disc boundary, while others search for subtle exudates. This parallel consensus creates true clinical-grade accuracy.
Cup-to-Disc Ratio anomaly detected in superior quadrant.
Clear vascular architecture. No exudates identified.
Signal noise in macula. Manual inspection required.
Chapter 03
The Final Clinical Verdict.
At the end of the pipeline, the refined vectors reach the MLP Decoder. Here, the abstract patterns are evaluated against 1.6 million clinical outcomes to produce the final pathology reading.
The noise is stripped away, leaving only the clinical truth—ready for physician review in milliseconds.
Built on RETFound.
The core diagnostic engine utilizes the world's first medical foundation model for ophthalmology. Published in Nature (2023) and developed by Moorfields Eye Hospital, it provides the robust feature extraction layer for all Clouseau screenings.