TreatmentFlow

TreatmentFlow is an AI-based system designed to optimize the flow of patients through the emergency room. It integrates three major AI components:

• Deep Learning-based Triage: Automatically assigns patient priority using a trained neural network.
• Constraint Optimization for Bed Assignment: Assigns hospital beds based on incoming patients and resource availability.
• Bayesian Network Diagnostics: Generates bedside documents for doctors, displaying the probabilities of specific conditions based on observed symptoms.

TreatmentFlow was originally developed by Adam Neto and Emese Elkind from February to April 2025 as a project for CISC 352 - Artificial Intelligence, a third-year course at Queen’s University School of Computing.

Usage instructions:

• Install all required dependencies (see below).

• Option 1: Run the TreatmentFlow_Lite.py file for access to our streamlined text-based UI for interacting with all 3 components of TreatmentFlow.

  • Deep Learning Component (uses DeepLearning/DLRaw.py)
    • Used to predict the a patient's CTAS priority level, whether or not they have performed a manual triage
  • Constraint Satisfaction Component (uses CSP/HospitalClasses.py and CSP/GreedyBedAssignment.py)
    • Used to assign patients to beds based on arrival time and priority
    • Also used to print the bed assignment through minute-by-minute simulated updates
  • Bayesian Network Component (uses BayesNets/BNRaw.py)
    • Used to generate patient diagnostic information based on the triage inputs (manually added or randomized) of the survey system
    • Displays diagnostic information in the console
    • Generates a patient diagnostic document in the bedside_documents folder

• Option 2: Run each component independently

  • Deep Learning Component (DeepLearningMLP.py)
    • Generates the confusion matrix and displays model accuracy for patient priority.
  • Constraint Satisfaction Component (MIPBedAssignment.py)
    • Compares solutions from both Mixed-Integer Programming (MIP) and a Greedy algorithm
    • To test Greedy-only scheduling:
      1. Create a hospital using the HospitalRecords class with at least 1 bed.
      2. Generate patients using the gen_patient_list() method.
      3. Initialize a scheduler with the Schedule class using the hospital object.
      4. Run run_hospital() and optionally waiting_times() to view Greedy algorithm results.
  • Bayesian Network Component (BayesSKlearn.py)
    • Generates the confusion matrix and displays model accuracy for diagnosis prediction.
    • Takes symptom inputs and outputs likely diagnoses and related probabilities.

Installation Instructions

Install the following Python packages before running the project:

python -m pip install ortools
pip install pandas pyreadr seaborn scikit-learn pymc3 matplotlib