Related Experiment Video
Updated: Jan 15, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
Medical Image Segmentation Assisted with Clinical Inputs via Language Encoder in A Deep Learning Framework.
Hengrui Zhao1, Biling Wang1, Deepkumar Mistry1
1Medical Artificial Intelligence and Automation (MAIA) Laboratory and Department of Radiation Oncology, University of Texas Southwestern Medical Center, Dallas, TX 75390, USA.
Machine Learning: Science and Technology
|October 13, 2025
Summary
This study introduces a deep learning framework for auto-segmentation in radiotherapy. It integrates clinical text data with medical images, significantly improving the accuracy of delineating clinical target volumes (CTVs) and organs at risk (OARs).
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiotherapy
Background:
- Accurate auto-segmentation of tumor volumes and organs at risk (OARs) is crucial for radiotherapy planning.
- Current auto-segmentation methods struggle with clinical target volumes (CTVs) due to limitations in image-based information.
Purpose of the Study:
- To develop a deep learning framework for medical image segmentation that integrates textual clinical information.
- To improve the accuracy of auto-segmentation for CTVs and OARs in radiotherapy treatment planning.
Main Methods:
- A transformer-based text encoder converts clinical data into vectors.
- These vectors are integrated with image features for enhanced segmentation.
- The framework was evaluated using prostate segmentation for localized prostate cancer radiation therapy.
Main Results:
- The proposed method significantly outperforms baseline and state-of-the-art methods.
- Integration of clinical context notably impacts delineation accuracy.
- The framework demonstrates high performance even with limited data.
Conclusions:
- The developed segmentation framework substantially enhances auto-segmentation accuracy for CTVs in cancer radiotherapy.
- This approach bridges the gap between automated methods and clinician-guided segmentation.
