Related Experiment Video
Updated: Jul 12, 2026
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
MDAL: Modality-difference-based active learning for multimodal medical image analysis via contrastive learning and
Haoran Wang1, Qiuye Jin1, Xiaofei Du1
1Digital Medical Research Center, School of Basic Medical Sciences, Fudan University, Shanghai 200032, China; Shanghai Key Laboratory of Medical Image Computing and Computer Assisted Intervention, Shanghai 200032, China.
This study introduces a new active learning framework, Multimodal Deep Active Learning (MDAL), to reduce annotation costs for multimodal medical images. MDAL effectively identifies informative samples, significantly lowering the need for labeled data in medical image analysis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Machine Learning
Background:
- Multimodal medical images offer complementary information for diagnostics.
- Deep learning requires large labeled datasets, but annotation is costly, especially for multimodal data.
- Existing active learning methods struggle with the complexities of multimodal medical image annotation.
Purpose of the Study:
- To develop a novel active learning framework to minimize annotation costs for multimodal medical image analysis.
- To quantify and leverage modality differences for efficient sample selection.
- To enable one-shot informative sample selection without initial labeled data.
Main Methods:
- Proposed a Multimodal Deep Active Learning (MDAL) framework.
- Quantified sample-wise modality differences using pointwise mutual information via multimodal contrastive learning.
- Introduced two sampling strategies: MaxMD and DiverseMD, based on modality differences.
Main Results:
- MDAL significantly outperforms advanced active learning competitors on brain glioma, meningioma, and ovarian cancer datasets.
- Achieved 99.6%, 99.9%, and 99.3% of fully supervised performance using only 20%, 20%, and 15% labeled samples, respectively.
- Demonstrated the ability to select informative samples in one shot without initial labeled data.
Conclusions:
- MDAL effectively reduces annotation costs in multimodal medical image analysis.
- The framework shows significant potential for broader application in multimodal medical data analysis.
- MDAL offers a cost-effective solution for leveraging deep learning in medical imaging.
