Live Platform
QML ASD
Analysis Platform
Analysis Platform
Explainable quantum machine learning for autism spectrum screening — from black box to glass box.
Accuracy
61.1%
↑ VQC model
F1 Score
0.607
Weighted avg
Analyses Run
175
Stored in Supabase
Qubits
4
2 ansatz layers · 24 params
Training Loss — VQC
40 epochs · Adam optimiser
Class Distribution
Test set · 175 samples
System Activity Log
API + DB events
Quantum State Amplitudes
Sample #42 — |ψ⟩ post-circuit
Run Pipeline
New Patient
Analysis
Analysis
Enter Q-CHAT-10 scores and demographic data. The VQC model will generate a prediction and store it in Supabase.
Input
Q-CHAT + Demo
→
Feature Ext.
Normalize
→
QML Model
VQC inference
→
Explainability
SHAP + LIME
→
Supabase
Store results
→
Visualize
Dashboard
Q-CHAT-10 Scores
Demographics
Analysis Result
No analysis yet.
Fill form and click Run.
Fill form and click Run.
Glass-Box AI
Explainability
Module
Module
SHAP feature attribution, sensitivity analysis, and LIME local explanations — making quantum predictions interpretable.
SHAP Feature Importance
mean |SHAP| over 60 test samples
Sensitivity Analysis
Mean |ΔP(ASD)| per ±0.15σ shift
LIME Waterfall — Sample #0
Local linear explanation · base P=0.45
Probability Distribution
ASD vs Non-ASD predicted probs
Glass-Box Interpretation Report
Key Finding 1
Behavioural Q-CHAT scores (A1–A10) account for over 80% of total SHAP attribution. The model has learned to weight diagnostic screening items correctly.
Key Finding 2
A2_Score (social attention) is the single most influential feature (mean |SHAP| = 0.094), consistent with clinical ASD literature on joint attention deficits.
Key Finding 3
Demographic features contribute minimally — the quantum model does not appear to exhibit demographic bias in its decision pathway.
Supabase Integration
Prediction
Database
Database
Every model inference is logged to Supabase — enabling audit trails, pattern analysis, and clinical oversight.
predictions table
supabase.io · public schema
Live inserts
| id | timestamp | patient_ref | prediction | confidence | top_feature | shap_score |
|---|
Schema Definition
CREATE TABLE predictions ( id uuid DEFAULT gen_random_uuid(), timestamp timestamptz DEFAULT now(), patient_ref text, features jsonb, prediction text, confidence float, shap_values jsonb, lime_coefs jsonb, top_feature text, shap_score float );
Python Integration
from supabase import create_client supabase = create_client(URL, KEY) # After model inference: supabase.table("predictions").insert({ "prediction": "ASD", "confidence": 0.87, "shap_values": shap_dict, "top_feature": "A2_Score", }).execute()
Prediction Distribution (DB)
Confidence Over Time
System Design
Platform
Architecture
Architecture
Full-stack AI platform: React frontend → FastAPI backend → PennyLane QML → Supabase storage.
System Architecture — NeuroLens QML Platform
👤 User / Clinician
browser
HTTPS
⚛ React Frontend
this dashboard
↓
🚀 FastAPI Backend
/predict · /explain · /store
/predict · /explain · /store
↓
⚛ PennyLane QML
VQC · SHAP · LIME
VQC · SHAP · LIME
quantum layer
results
🗄 Supabase
PostgreSQL · REST · Realtime
PostgreSQL · REST · Realtime
cloud database
Frontend
API Layer
QML Engine
Database
VQC Circuit Diagram
|0⟩ q₀
Ry(x₀)
Rz·Ry·Rz
CNOT
Rz·Ry·Rz
CNOT
⟨Z₀⟩
|0⟩ q₁
Ry(x₁)
Rz·Ry·Rz
CNOT
Rz·Ry·Rz
CNOT
|0⟩ q₂
Ry(x₂)
Rz·Ry·Rz
CNOT
Rz·Ry·Rz
CNOT
|0⟩ q₃
Ry(x₃)
Rz·Ry·Rz
CNOT
Rz·Ry·Rz
CNOT
Encode: AngleEmbedding
Ansatz: StronglyEntangling
Measure: ⟨PauliZ⟩
Tech Stack
Frontend
HTML / CSS / JS + Chart.js
Backend API
FastAPI + Uvicorn
Quantum ML
PennyLane + NumPy VQC
Explainability
SHAP + LIME + Sensitivity
Database
Supabase (PostgreSQL)
Dataset
UCI ASD Screening