Related Experiment Video
Updated: Jul 16, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.8K
Foundation Models for Quantitative Biomarker Discovery in Cancer Imaging
Suraj Pai1,2,3, Dennis Bontempi1,2,3, Vasco Prudente1,2,3
1Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Harvard Institutes of Medicine, 77 Avenue Louis Pasteur, Boston, MA 02115, United States of America.
Medrxiv : the Preprint Server for Health Sciences
|September 21, 2023
Summary
Foundation models, trained on vast data, accelerate imaging biomarker discovery in medicine. These models improve efficiency and performance, especially with limited data, outperforming traditional methods.
Area of Science:
- Artificial Intelligence
- Medical Imaging
- Biomarker Discovery
Background:
- Foundation models offer a powerful paradigm in deep learning, leveraging large-scale self-supervised pre-training.
- Medical applications often face challenges due to scarce labeled data for training specialized models.
- Imaging biomarkers are crucial for disease diagnosis and monitoring, but their discovery can be data-intensive.
Approach:
- Developed a foundation model using a convolutional encoder trained via self-supervised learning on 11,467 radiographic lesions.
- Evaluated the model's efficacy in discovering imaging biomarkers for distinct, clinically relevant applications.
- Compared the performance of foundation model-derived biomarkers against conventional supervised learning approaches.
Key Points:
- Foundation models significantly enhance the efficiency and performance of imaging biomarker discovery.
- Models derived from foundation models outperformed traditional supervised models, particularly with limited training data.
- Foundation models demonstrated increased stability against input variations and stronger biological associations.
Conclusions:
- Foundation models hold immense potential for discovering novel imaging biomarkers.
- This approach can accelerate the translation of imaging biomarkers into clinical practice.
- The methodology shows promise for broader applications in medical AI and clinical use cases.
