Update README.md

README.md

@@ -5,28 +5,65 @@ emoji: 🩸
 colorFrom: red
 colorTo: gray
 sdk: gradio
-sdk_version: "4.0"
+sdk_version: 4.44.1
 app_file: app.py
 pinned: false
+tags:
+- medical
+- biology
+- xai
+- malaria
+- sickle-cell
 ---
 
-# Explainable AI-Driven Ensemble Model for Malaria and Sickle Cell Anemia Classification in Western Kenya
+# Hemaclass AI Diagnostics: Clinical Decision Support System
 
-## Project Overview
-The **Hemaclass AI Diagnostics** dashboard is a Phase 4 prototype designed to classify multi-class hematological diseases (Malaria, Sickle Cell Anemia, Co-infections, and Negative cases) using a highly optimized, stacked ensemble machine learning architecture.
+An explainable ensemble-based diagnostic tool optimized for **Malaria**, **Sickle Cell Anemia (SCA)**, and **Co-infection** classification in Western Kenya.
 
-## Clinical Decision Support
-The tool is designed as an assistive diagnostic dashboard. Clinicians can input manual patient vitals and lab results (e.g., Hemoglobin, WBC count, Malaria RDT) or batch upload retrospective `.csv` clinical data to receive:
-1. **AI Diagnosis:** Fast classification of the patient's condition.
-2. **Confidence Score:** Probabilistic output from the meta-learner.
-3. **Clinical Recommendations:** Automated WHO-aligned protocol suggestions based on severity (e.g., hyperhemolytic crisis alerts).
-4. **XAI Interpretability:** Local SHAP (SHapley Additive exPlanations) waterfall plots explaining exactly *why* the AI made the decision.
+## 🔬 Model Details
+- **Architecture:** Stacking Ensemble (RF, SVM, XGBoost) with a Logistic Regression Meta-Learner.
+- **Explainability:** Integrated SHAP waterfall plots for local feature importance.
+- **Preprocessing:** MICE Imputation, SMOTE balancing, and Z-Score Normalization.
 
-## Model Architecture
-- **Base Learners:** Random Forest, Support Vector Machine (RBF Kernel), XGBoost.
-- **Meta-Learner:** Logistic Regression (Stacking Classifier).
-- **Preprocessing:** MICE Imputation, SMOTE Data Balancing, robust Z-Score Normalization.
-- **Optimization:** ~30 minute Bayesian Search CV (Nested 5-Fold) to guarantee robust generalization without overfitting.
+## 🚀 Quick Usage (Inference Code)
+To use this model programmatically, ensure you have your `.pkl` artifacts in the local directory.
 
-## Ethical & Regulatory Compliance
-This model assumes data has been ethically sourced, anonymized, and cleaned in compliance with the Kenya Data Protection Act (2019). It is positioned strictly as a decision-support tool. **The final diagnostic authority always remains with the attending clinician.**
+```python
+import joblib
+import pandas as pd
+import numpy as np
+
+# 1. Load Artifacts
+model = joblib.load('ensemble_model.pkl')
+scaler = joblib.load('scaler.pkl')
+imputer = joblib.load('imputer.pkl')
+FEATURES = joblib.load('feature_names.pkl')
+target_names = ['Negative', 'Malaria', 'SCA', 'Co-infection']
+
+def predict_patient(data_dict):
+    """
+    Input: Dictionary of patient vitals/labs
+    Output: Predicted Diagnosis and Confidence
+    """
+    # Create DataFrame and align features
+    df = pd.DataFrame([data_dict])
+    for col in set(FEATURES) - set(df.columns):
+        df[col] = np.nan
+    df = df[FEATURES]
+
+    # Preprocess
+    X_imp = imputer.transform(df)
+    X_scaled = scaler.transform(X_imp)
+
+    # Inference
+    pred = model.predict(X_scaled)
+    prob = np.max(model.predict_proba(X_scaled))
+    
+    return {
+        "Diagnosis": target_names[pred],
+        "Confidence": f"{prob*100:.2f}%"
+    }
+
+# Example Usage:
+patient_data = {'age': 25, 'hb': 10.5, 'temp': 38.5, 'malaria_rdt': 1}
+print(predict_patient(patient_data))