Password Check
The frontend sends the shared password to
/api/login. The backend verifies it and returns a
short-lived signed token.
SECURE SCREENING ACCESS
Password-protected image screening with backend-only model inference.
Enter the shared access password, upload one face image, and receive a simple screening summary generated by the FastAPI backend. The Keras model stays on the server and never reaches the browser.
Shared password only
Short-lived token
5 MB upload limit
JPG / PNG only
WORKFLOW
Two simple steps, with the model always staying on the backend.
Image Upload
After login, the frontend uploads one JPG or PNG image to
/api/predict with the bearer token attached.
Backend Inference
FastAPI validates the file, rate-limits the request, and runs the Keras model in memory without exposing the model file to the browser.
Readable Summary
The UI shows a simplified follow-up indication, signal comparison, and a reminder that this result supports screening rather than formal diagnosis.
STEP 1
Unlock access
The password is verified on the server only and is never embedded in the frontend.
Session token has not been issued yet.
STEP 2
Upload image
Upload one face image. The backend validates file type and size before running inference.
Login is required before uploads are enabled.
RESULT
Explainable screening output designed for parent and clinical review.
Latest screening outcome
Awaiting result
Predicted category: --
Awaiting image
Upload an image to receive a screening summary.
Autism category probability
--
The threshold comparison will appear here after screening.
Probability comparison
How to interpret this result
This is a screening support result and should be read alongside clinical judgement, developmental history, and follow-up assessment.
Interpretation guide
Use the output as a screening aid for follow-up, not as a diagnosis.
Compare the two categories
The visual bars show how the current image leans between the Autism and Non-autism categories.
Read it with context
Image-based screening should be interpreted together with observation, history, and formal clinical assessment.
Treat it as follow-up support
The goal is to help flag whether closer follow-up may be helpful, not to confirm or rule out diagnosis on its own.
SECURITY
Minimal controls that keep the workflow simple and safer.
Backend-only model
model_asian.keras stays on the server and is loaded
once at startup.
Shared password
The password is stored in backend environment variables, never in browser code.
Short-lived token
Browser sessions use signed bearer tokens that expire quickly and
live only in sessionStorage.
Input guardrails
The backend accepts only JPG and PNG uploads, limits file size, generates safe filenames, and applies simple rate limits.