Related Experiment Video
Updated: May 27, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
SegFormer3D: Improving the Robustness of Deep Learning Model-Based Image Segmentation in Ultrasound Volumes of the
Benjamin Hers1, Maria Bonta2, Siyi Du1
1Department of Electrical and Computer Engineering, University of British Columbia, Vancouver, British Columbia, Canada.
Insights
Developmental dysplasia of the hip (DDH) diagnosis can be improved with new 3D ultrasound and AI. Our enhanced SegFormer model offers greater accuracy and robustness for pediatric orthopedic imaging.
Area of Science:
- Orthopedics
- Medical Imaging
- Artificial Intelligence
Background:
- Developmental dysplasia of the hip (DDH) affects 1-3% of newborns, potentially causing long-term disability if untreated.
- Current diagnosis relies on manual 2D ultrasound measurements, which are time-consuming and prone to errors.
- Deep learning with 3D ultrasound shows promise for improving DDH metric accuracy but lacks robustness for clinical use.
Purpose of the Study:
- To quantitatively evaluate the robustness of current state-of-the-art 3D ultrasound bone segmentation for DDH.
- To propose and validate a novel 3D deep learning model for improved accuracy and robustness in DDH diagnosis.
Main Methods:
- Evaluation of convolutional neural network and vision transformer models for 3D ultrasound bone segmentation robustness.
- Development of a 3D extension of the SegFormer architecture with hierarchically structured encoders.
- Quantitative assessment of the proposed model on clinical pediatric patient data, including variations in resolution and anatomy.
Main Results:
- Identified limitations in current deep learning models, showing failures under common data variations.
- The proposed 3D SegFormer extension demonstrated improved accuracy and robustness.
- Achieved up to 0.9% higher Dice score and 3% smaller Hausdorff distance 95% compared to state-of-the-art methods on unseen data variations.
Conclusions:
- Current automated 3D ultrasound segmentation for DDH requires enhanced robustness for clinical deployment.
- The proposed 3D SegFormer model offers a more accurate and reliable approach for diagnosing DDH using 3D ultrasound.
- This advancement has the potential to improve diagnostic efficiency and patient outcomes in pediatric orthopedics.
Abstract:
Developmental dysplasia of the hip (DDH) is a painful orthopedic malformation diagnosed at birth in 1-3% of all newborns. Left untreated, DDH can lead to significant morbidity including long-term disability. Currently the condition is clinically diagnosed using 2-D ultrasound (US) imaging acquired between 0 and 6 mo of age. DDH metrics are manually extracted by highly trained radiologists through manual measurements of relevant anatomy from the 2-D US data, which remains a time-consuming and highly error-prone process. Recently, it was shown that combining 3-D US imaging with deep learning-based automated diagnostic tools may significantly improve accuracy and reduce variability in measuring DDH metrics. However, the robustness of current techniques remains insufficient for reliable deployment into real-life clinical workflows. In this work, we first present a quantitative robustness evaluation of the state of the art in bone segmentation from 3-D US and demonstrate examples of failed or implausible segmentations with convolutional neural network and vision transformer models under common data variations, e.g., small changes in image resolution or anatomical field of view from those encountered in the training data. Second, we propose a 3-D extension of SegFormer architecture, a lightweight transformer-based model with hierarchically structured encoders producing multi-scale features, which we show to concurrently improve accuracy and robustness. Quantitative results on clinical data from pediatric patients in the test set showed up to 0.9% improvement in Dice score and up to a 3% smaller Hausdorff distance 95% compared with state of the art when unseen variations in anatomical structures and data resolutions were introduced.
