Related Experiment Video
Updated: Sep 21, 2025
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
42.9K
A Multi-Task Cross-Task Learning Architecture for Ad Hoc Uncertainty Estimation in 3D Cardiac MRI Image Segmentation
S M Kamrul Hasan1, Cristian A Linte1,2
1Chester F. Carlson Center for Imaging Science, Rochester, NY.
Computing in Cardiology
|June 1, 2022
Summary
This study introduces a novel multi-task learning method for accurate 3D cardiac MRI segmentation. The approach leverages unlabeled data and uncertainty estimation to improve segmentation mask quality and reliability.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiology
Background:
- Deep learning significantly enhances medical image segmentation.
- Semi-supervised learning (SSL) is crucial for improving model performance using unlabeled data.
- Multi-task learning boosts model generalizability.
Purpose of the Study:
- To develop a robust segmentation method for 3D cardiac MR images.
- To improve segmentation accuracy and smoothness using a novel approach.
- To integrate uncertainty estimation for reliable failure detection.
Main Methods:
- A Multi-task Cross-task learning consistency approach was proposed.
- Enforced correlation between pixel-level segmentation and geometric-level distance map tasks.
- Utilized varying quantities of labeled data and Convolutional Neural Networks (CNNs).
Main Results:
- Demonstrated the model's effectiveness in segmenting the left atrial cavity from GE-MR images.
- Validated the approach across diverse labeled data quantities.
- Successfully incorporated uncertainty estimates to identify low-quality segmentation masks.
Conclusions:
- The proposed method effectively segments 3D cardiac MR images.
- The model shows potential for flagging unreliable segmentation outputs.
- This approach enhances the reliability of automated cardiac image analysis.
