|
|
| """
|
| Example: Cardiotoxicity Risk Prediction
|
| Demonstrates how to use the deployed model for individual patient prediction
|
| """
|
|
|
| import pickle
|
| import pandas as pd
|
| import json
|
|
|
|
|
| print("Loading model and pipelines...")
|
| with open('cardiotoxicity_model.pkl', 'rb') as f:
|
| model = pickle.load(f)
|
|
|
| with open('preprocessing_pipeline.pkl', 'rb') as f:
|
| preprocessing = pickle.load(f)
|
|
|
| with open('feature_engineering_pipeline.pkl', 'rb') as f:
|
| feature_engineering = pickle.load(f)
|
|
|
| with open('model_metadata.json', 'r') as f:
|
| metadata = json.load(f)
|
|
|
| selected_features = metadata['features']['selected_features']
|
| print("✓ Model loaded successfully!\n")
|
|
|
|
|
| print("="*80)
|
| print("EXAMPLE PATIENT PREDICTIONS")
|
| print("="*80)
|
|
|
|
|
| patient_1 = pd.DataFrame({
|
| 'age': [45],
|
| 'weight': [65],
|
| 'height': [160],
|
| 'heart_rate': [70],
|
| 'LVEF': [65],
|
| 'PWT': [0.8],
|
| 'LAd': [3.2],
|
| 'LVDd': [4.5],
|
| 'LVSd': [2.8],
|
| 'AC': [0],
|
| 'antiHER2': [1],
|
| 'HTA': [0],
|
| 'DL': [0],
|
| 'smoker': [0],
|
| 'exsmoker': [0],
|
| 'diabetes': [0],
|
| 'obesity': [0],
|
| 'ACprev': [0],
|
| 'RTprev': [0],
|
| 'heart_rhythm': [0]
|
| })
|
|
|
|
|
| patient_2 = pd.DataFrame({
|
| 'age': [62],
|
| 'weight': [80],
|
| 'height': [165],
|
| 'heart_rate': [88],
|
| 'LVEF': [52],
|
| 'PWT': [1.1],
|
| 'LAd': [4.2],
|
| 'LVDd': [5.5],
|
| 'LVSd': [3.8],
|
| 'AC': [1],
|
| 'antiHER2': [1],
|
| 'HTA': [1],
|
| 'DL': [1],
|
| 'smoker': [0],
|
| 'exsmoker': [1],
|
| 'diabetes': [1],
|
| 'obesity': [1],
|
| 'ACprev': [1],
|
| 'RTprev': [1],
|
| 'heart_rhythm': [0]
|
| })
|
|
|
| def predict_patient(patient_data, patient_name):
|
| """Predict CTRCD risk for a patient"""
|
| print(f"\n{patient_name}:")
|
| print("-" * 80)
|
|
|
|
|
| is_valid, message = preprocessing.validate(patient_data)
|
| if not is_valid:
|
| print(f"❌ Validation failed: {message}")
|
| return
|
|
|
|
|
| processed = preprocessing.preprocess(patient_data)
|
| engineered = feature_engineering.engineer_features(processed)
|
| X = engineered[selected_features]
|
|
|
|
|
| risk_prob = model.predict_proba(X)[:, 1][0]
|
|
|
|
|
| if risk_prob < 0.3:
|
| risk_level = "LOW"
|
| recommendation = "Standard monitoring (every 6 months)"
|
| elif risk_prob < 0.5:
|
| risk_level = "MODERATE"
|
| recommendation = "Enhanced monitoring (every 3 months)"
|
| elif risk_prob < 0.7:
|
| risk_level = "HIGH"
|
| recommendation = "Cardioprotective agents + frequent monitoring"
|
| else:
|
| risk_level = "VERY HIGH"
|
| recommendation = "Consider treatment modification + cardiology consult"
|
|
|
| print(f" CTRCD Risk Probability: {risk_prob:.2%}")
|
| print(f" Risk Level: {risk_level}")
|
| print(f" Clinical Recommendation: {recommendation}")
|
|
|
|
|
| print(f"\n Key Clinical Parameters:")
|
| print(f" - Age: {patient_data['age'].iloc[0]} years")
|
| print(f" - LVEF: {patient_data['LVEF'].iloc[0]}%")
|
| print(f" - Heart Rate: {patient_data['heart_rate'].iloc[0]} bpm")
|
|
|
| cv_risk_count = sum([
|
| patient_data['HTA'].iloc[0],
|
| patient_data['DL'].iloc[0],
|
| patient_data['smoker'].iloc[0] or patient_data['exsmoker'].iloc[0],
|
| patient_data['diabetes'].iloc[0],
|
| patient_data['obesity'].iloc[0]
|
| ])
|
| print(f" - CV Risk Factors: {cv_risk_count}")
|
|
|
| treatment_burden = sum([
|
| patient_data['AC'].iloc[0],
|
| patient_data['antiHER2'].iloc[0],
|
| patient_data['ACprev'].iloc[0],
|
| patient_data['RTprev'].iloc[0]
|
| ])
|
| print(f" - Treatment Burden: {treatment_burden}")
|
|
|
|
|
| predict_patient(patient_1, "PATIENT 1: Low Risk Profile")
|
| predict_patient(patient_2, "PATIENT 2: High Risk Profile")
|
|
|
| print("\n" + "="*80)
|
| print("Predictions completed successfully!")
|
| print("="*80)
|
|
|
