ZhanAI

Clinical Decision Support System

Myopia Screening Assessment

Enter patient demographics and upload a retinal fundus photograph. The multimodal AI model (SwinV2-Tiny + MLP fusion, trained on ODIR-5K, val AUC 0.946) will classify the eye as “Myopia” or “No Myopia” with a confidence score.

Patient Clinical Parameters

All fields are required. Values feed the tabular MLP branch of the fusion model.

ODIR-5K cohort range: 1–110 years

Negative = myopic  ·  step 0.25 D

Normal range: 22–27 mm  ·  step 0.01 mm

Fundus Photograph

Upload a retinal fundus image (JPEG or PNG, max 20 MB). The SwinV2-Tiny vision branch expects a 256 × 256 input — resizing is handled automatically.

Drop fundus image here

or browse files · JPEG, PNG · max 20 MB

No analysis performed yet

Fill in the patient parameters, upload a fundus image, and click “Run Progression Analysis”.

Model Architecture

  • DatasetODIR-5K (6392 fundus images)
  • Vision branchSwinV2-Tiny (256×256 → 768-d)
  • Tabular branchMLP: 2 → 128 → 64
  • FusionConcat 832 → Linear 256 → 2
  • ClassesNo Myopia · Myopia
  • Val AUC0.9463 (epoch 4)
  • LossFocal Loss (γ=2, α=0.25)
  • PrecisionMixed-precision (AMP)