The Latent
Neural Ingress
How Clouseau transforms raw clinical pixels into multi-dimensional latent vectors using state-of-the-art transformer backbones.
Real-time Vector Space Mapping
RETFound Backbone
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.
- ✚ Trained on 1.6 Million Retinal Scans
- ✚ Masked Autoencoder (MAE) Pre-training
- ✚ Transferrable to Cardiac & Stroke risk
SegFormer CDR Engine
For structural Cup-to-Disc analysis, Clouseau employs a specialized hierarchical Transformer (SegFormer) trained on the REFUGE open-source clinical dataset.
- ✚ All-MLP Decoder Architecture
- ✚ Resolution-Invariant Inference
- ✚ Multi-level Pathological Feature Maps
The Geometric Constant
Our analysis pipeline doesn't just "see" an image; it projects pixels into a high-dimensional Latent Space where semantic relationships (like the border between healthy tissue and a lesion) are mathematically isolated.
By stripping away clinical noise through self-supervised learning, the models lock onto the core biological indicators required for critical triaging.
> Input: shape(3, 224, 224)
> Extracting Patches... done.
> Linear Projection [Vector Dim: 1024]
> MSA_Attention_Head_01: Active [0.88]
> Final MLP Cls: "PATHOLOGY_DETECTED"