Optimal Biopsy Decision Making for Breast Cancer using Reinforcement Learning

This repository contains the implementation of the project "Optimal Biopsy Decision Making for Breast Cancer using Reinforcement Learning (RL)." The project explores two primary RL methods: Backward Induction and Q-Learning, with an aim to generate optimal biopsy decision policies for different stages of breast cancer.

Project Phases

1. Initial Setup

Implemented a Backward Induction algorithm as referenced in the literature review to generate policies for various stages (probability of cancer) at each timestamp (age), starting from 40 to 100 years.

2. Q-Learning

Implemented an advanced RL method, Q-Learning, to generate policies for each age-stage combination and compared the results with those from Backward Induction.

3. Testing Both Policies

Tested both policies via two methods:

• Simulation Testing: Simulated scenarios to evaluate the performance of the policies.
• RSNA Screening Data: Tested policies on RSNA screening data from Kaggle.

4. Generating Granular Policies

Generated policies at a more granular level with risk scores ranging from 0 to 100 in increments of 0.1 using both RL methods to observe improvements.

Simulation Testing Details

In the simulation testing phase, policies were evaluated based on their performance in simulated environments. Mean and standard deviation (std) of rewards were calculated to measure the policies' effectiveness and stability.

Results

Descriptive statistics and detailed data analysis of the working dataset were provided. The results section covers the comprehensive performance of the implemented RL methods, including mean rewards and standard deviations.

Discussion

The discussion section highlights the scope for future research, particularly focusing on DQN techniques. The potential for using large clinical datasets and CAD models employed by radiologists for accurate testing and results is also discussed.

Getting Started

Prerequisites

• Python 3.x
• Required libraries: numpy, pandas, etc.

Installation

Clone the repository to your local machine:

git clone /~https://github.com/ruchithakor/MRP_Optimal_Biopsy_Decision_Making_Breast_Cancer_RL.git
cd MRP_Optimal_Biopsy_Decision_Making_Breast_Cancer_RL

Running the Code

Navigate to the directory MRP_Optimal_Biopsy_Decision_Making_Breast_Cancer_RL and run the scripts. Example given below for running backward Induction file:

cd MRP_Optimal_Biopsy_Decision_Making_Breast_Cancer_RL
python code/backward_induction.py
# for granular level
python code/granular_level_implementation/backward_induction_granular.py

Author

Ruchi Parmar