Related Experiment Video
Updated: Jan 15, 2026
04:25
Author Spotlight: Bridging Gaps in Anatomy and Establishing a Foundation for Algorithmic Studies
Published on: December 15, 2023
3.7K
Deep Unsupervised Clustering for Prostate Auto-segmentation With and Without Hydrogel Spacer
Hengrui Zhao1, Biling Wang1, Michael Dohopolski1
1Medical Artificial Intelligence and Automation (MAIA) Laboratory and Department of Radiation Oncology, University of Texas Southwestern Medical Center, Dallas, TX, U.S.A.
Machine Learning: Science and Technology
|October 13, 2025
Summary
Deep learning models struggle with heterogeneous clinical data. This study introduces CLIP-UNet, a text-guided segmentation model that uses clustering to improve prostate cancer radiotherapy auto-segmentation accuracy.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiotherapy
Background:
- Clinical datasets for deep learning (DL) models are often heterogeneous.
- Hydrogel spacers used in prostate cancer radiotherapy alter CT appearance, reducing auto-segmentation accuracy.
Purpose of the Study:
- To address heterogeneity in clinical datasets for prostate cancer radiotherapy.
- To improve auto-segmentation accuracy using a novel clustering and text-guided segmentation approach.
Main Methods:
- Collected a clinical dataset of 909 patients with and without hydrogel spacers.
- Used Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction and k-means clustering.
- Developed CLIP-UNet, a text-guided segmentation model encoding cluster information for improved segmentation.
Main Results:
- UMAP identified up to three distinct clusters within the dataset.
- CLIP-UNet achieved a Dice score of 86.2%, outperforming the baseline UNet (84.4%).
- CLIP-UNet demonstrated superior performance compared to other state-of-the-art models.
Conclusions:
- Deep learning-assisted automatic clustering can uncover hidden data structures in clinical datasets.
- CLIP-UNet effectively leverages clustered labels to achieve higher segmentation performance in prostate cancer radiotherapy.
