Related Experiment Video
Updated: Jul 4, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
Custom-Trained Deep Learning-Based Auto-Segmentation for Male Pelvic Iterative CBCT on C-Arm Linear Accelerators.
Riley C Tegtmeier1, Christopher J Kutyreff1, Jennifer L Smetanick1
1Department of Radiation Oncology, Mayo Clinic Arizona, Phoenix, Arizona.
Practical Radiation Oncology
|February 7, 2024
Summary
A new artificial intelligence tool accurately auto-segments prostate cancer treatment areas on iCBCT scans, improving efficiency and accuracy compared to older methods. While effective for intact prostates, prostate bed segmentation requires further refinement.
Area of Science:
- Radiotherapy and Medical Imaging
- Artificial Intelligence in Oncology
- Image Segmentation Techniques
Background:
- Accurate auto-segmentation of organs at risk and target volumes is crucial for effective radiotherapy planning.
- Deep learning auto-segmentation (DLAS) tools offer potential for improving contouring efficiency and consistency.
- Evaluating DLAS performance on enhanced iterative cone beam computed tomography (iCBCT) is essential for clinical applicability.
Purpose of the Study:
- To assess the clinical utility of a commercial artificial intelligence-driven DLAS tool for prostate and prostate bed treatments using iCBCT.
- To compare the performance of an iCBCT-trained DLAS model against a pCT-trained DLAS model and a deformable image registration (DIR) method.
Main Methods:
- DLAS models were trained on 116 iCBCT datasets with manual delineations of organs at risk (bladder, femoral heads, rectum) and target volumes (prostate, prostate bed).
- An independent set of 25 iCBCT datasets was used for model testing.
- Segmentation accuracy was evaluated using geometric metrics (Dice Similarity Coefficient, Hausdorff distance) and qualitative assessment by experts, compared to reference and other automated methods.
Main Results:
- The iCBCT-trained DLAS model significantly outperformed the pCT-trained DLAS model and DIR method for organs at risk and intact prostate segmentation, with mean DSC ≥0.90.
- Performance for prostate bed segmentation was suboptimal (mean DSC <0.75), with ~63% of contours requiring significant manual editing.
- Clinical acceptability scores were high for bladder (90%), femoral heads (93%), and intact prostate (92%), but lower for rectum (67%) and prostate bed.
Conclusions:
- The iCBCT-trained DLAS tool shows promise for enhancing segmentation accuracy and efficiency in iCBCT-based radiotherapy, outperforming DIR methods.
- Further optimization is needed for the DLAS tool's performance on prostate bed segmentation to achieve widespread clinical adoption.
- The study highlights the potential of AI-driven auto-segmentation for improving radiotherapy workflows in clinical practice.
