Researchers at CMITT are working on ways to update deep learning models for brain tumor segmentation so that they can adapt to changes in data. This is important because medical images are constantly changing, as new scanners and modalities are developed. This work will be presented at MICCAI 2023.
- Background: Deep learning models have been used to segment anatomical structures in medical images. However, these models can only perform well if they are trained on data from a single source domain. If the data changes, the model may not be able to adapt and will perform poorly.
- Objective: This study aimed to develop a method for updating deep learning models so that they can adapt to changes in data.
- Methods: The researchers developed a method using “incremental learning”. This method allows the model to be updated with new data without forgetting what it has already learned.
- Results: The researchers evaluated their method on a brain tumor segmentation task. They found that the method was able to well retain the discriminability of previously learned structures, even when the data was changed.
- Conclusion: The study’s findings suggest that incremental learning is a promising method for updating deep learning models. This method could be used to improve the performance of deep learning models in a variety of medical imaging tasks.
heterogeneous structure-incremental) and several other machine learning methods.
Link to arXiv paper: https://arxiv.org/abs/2305.19404
Xiaofeng Liu, Helen A. Shih, Fangxu Xing, Emiliano Santarnecchi, Georges El Fakhri, Jonghye Woo, “Incremental Learning for Heterogeneous Structure Segmentation in Brain Tumor MRI” arXiv paper: https://arxiv.org/abs/2305.19404
To be presented at MICCAI 2023