Ultrasound Image Segmentation and Reinforcement Learning Navigation

Project Overview

This project combines deep learning-based image segmentation with reinforcement learning to automatically navigate to regions of interest in ultrasound images. The system consists of three main components:

1. Image Segmentation: A ResNet-based U-Net model trained to segment regions of interest in abdominal ultrasound images.
2. Center Detection: An algorithm to find the centers of the segmented regions.
3. Reinforcement Learning Navigation: A DQN (Deep Q-Network) agent trained to navigate to the centers of the segmented regions.

Blog Posts

• Ultrasound Image Segmentation and Reinforcement Learning Navigation

Results

The trained agent successfully navigates to the centers of the segmented regions with a high success rate. The improvements to address oscillations significantly enhanced the agent's performance.

Segmentation Results

Navigation Example

Read the full blog post to see more results and details →

Repository

The complete code for this project is available on GitHub.