Connecting…
--:--:--

Clinical AI Dashboard

ML: LR + GaussianNB Ensemble · Dataset: Synthetic QA Benchmark v2.1 (N=120, 15 classes) · PubMed-inspired

120
Training Cases
15 disease classes
81.2%
ML Accuracy
LR+NB Ensemble
80.3%
Precision
Macro-averaged
77.8%
Recall
Macro-averaged
94.7%
Emergency Sens.
Safety-critical
3.1%
Undertriage
Below 5% target
0
Active Sessions
Live count
Per-Category ML Performance
Quick Actions
ML Pipeline: Input → Feature Extraction (12-dim) → LR+NB Ensemble → Calibration → 9-Section Output
Emergency Detection Matrix
36
True Positive
1
False Positive
2
False Negative
11
True Negative
Emergency sensitivity: 94.7% · Undertriage: 3.1%
ML Model Architecture
Primary ModelLogistic Regression
Secondary ModelGaussian Naive Bayes
Ensemble WeightLR 60% / NB 40%
Feature Dims12 binary/continuous
Disease Classes15 categories
Train/Test Split70% / 30%
Dataset Info

• Source: Synthetic Clinical QA Benchmark v2.1

• Inspiration: PubMed-style clinical QA pairs

• Size: N=120 cases, 15 disease classes

• Categories: Cardiac, Pulmonary, Neuro, Infectious, GI, Metabolic, Other

• Features: 12-dim symptom + vital sign vector

• Validation: Stratified 70/30 train-test split

FOR RESEARCH / DECISION SUPPORT ONLY

ModeHospital AI Assist
Patient ID—
StatusAwaiting Input
ML EngineLR + GaussianNB
READY

AI Clinical Diagnosis

9-section structured output · ML prediction · Differential diagnosis · Triage · Evaluation metrics

👤Patient Information
📊Vital Signs
🩺Symptoms
Quick Select
📋History & Medications
🔬
Clinical AI Output
Fill in patient data and click "Run Clinical AI"
to generate a full 9-section structured assessment

Patient Board

Live simulation · Priority triage queue · Haemodynamic monitoring

Loading patients…

Decision Console

Multi-agent clinical reasoning · Real-time grading · Risk analysis

Select Clinical Scenario
Loading…
Active Patient
🏥
No scenario selected
Select a task and click Reset Episode
Submit Clinical Decision
Select a task to begin
Grading Result + AI Reasoning
⚖️
Awaiting Decision

AI vs AI Benchmark

Human · ClinicalReasoningEngine (ML) · Rule-Based System

Benchmark Configuration
🤖
Ready to Benchmark
Select task and submit to compare AI systems

Model Evaluation Report

LR + GaussianNB Ensemble · Synthetic QA Benchmark v2.1 · N=120 cases · 15 disease classes

Loading evaluation metrics…

System Architecture

ML pipeline · Dataset · Feature engineering · Evaluation methodology

System Pipeline Flow
Patient Input
symptoms, vitals, demographics
→
Feature Extraction
12-dim binary/continuous
→
ML Ensemble
LR (60%) + NB (40%)
→
Probability Calibration
Platt scaling + history boost
→
Triage Classification
ML signal + vital override
→
Clinical Reasoning
6-step trace engine
→
9-Section Output
structured clinical report
Feature Engineering (12 Dimensions)
Disease Classes (15)
Dataset Description

Name: Synthetic Clinical QA Benchmark v2.1

Inspiration: PubMed clinical QA pairs

Total cases: 120 (8 per disease class)

Disease classes: 15 across 7 categories

Generation: Prototype vectors + Gaussian noise (σ=0.15)

Split: 70/30 train-test, stratified by class

Not real patient data. Research use only.

ML Architecture Details

Model 1: Logistic Regression (C=1.5, multinomial, lbfgs)

Model 2: Gaussian Naive Bayes (var_smoothing=1e-8)

Ensemble: Weighted average (LR 60% + NB 40%)

Scaling: StandardScaler (zero-mean, unit-variance)

Calibration: History-based Bayesian prior adjustment

Override: Vital sign triage escalation logic

Fallback: Cosine similarity (sklearn unavailable)

Evaluation Methodology

Split: 70/30 stratified train-test

Accuracy: Macro-averaged across 15 classes

Precision: Macro-averaged (zero_division=0)

Recall: Macro-averaged (zero_division=0)

F1: Harmonic mean of precision and recall

Emergency: Binary sensitivity/specificity

Undertriage: FN rate for emergency cases