Automated Contouring Using Neural Networks

This white paper presents the Contour ProtégéAI+™ framework, which utilizes neural networks for the automated segmentation of structures in medical images, specifically CT and MR images. The document outlines the challenges associated with manual segmentation and the limitations of traditional algorithms. It details the architecture of the neural network model based on U-Net, explaining how it processes input images to generate segmentation masks. The training and validation processes are described, including the use of large, multi-institution datasets to ensure robust performance. The paper presents various metrics used to evaluate the model's effectiveness, such as the Dice coefficient and mean distance to agreement, alongside qualitative user feedback. Results indicate that the neural network's segmentation performance is comparable to or superior to existing atlas-based methods. The findings are supported by tables and figures that illustrate the model's performance across different anatomical structures, emphasizing its potential for improving efficiency in clinical practice.