Related Experiment Video
Updated: Jun 29, 2025
06:51
Utilizing 18F-FDG PET/CT Imaging and Quantitative Histology to Measure Dynamic Changes in the Glucose Metabolism in Mouse Models of Lung Cancer
Published on: July 21, 2018
17.9K
Foundation model for cancer imaging biomarkers
Suraj Pai1,2,3, Dennis Bontempi1,2,3, Ibrahim Hadzic1,2,3
1Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Harvard Institutes of Medicine, Boston, MA USA.
Nature Machine Intelligence
|March 25, 2024
Summary
Foundation models, trained on extensive data, accelerate cancer imaging biomarker discovery. This approach reduces the need for large labeled datasets, outperforming traditional methods in clinical applications.
Area of Science:
- Artificial Intelligence
- Deep Learning
- Medical Imaging
Background:
- Foundation models are large-scale AI models trained on vast datasets for diverse applications.
- Self-supervised learning (SSL) is key to foundation models, reducing reliance on labeled data.
- Scarcity of labeled medical data hinders AI development in healthcare.
Purpose of the Study:
- To develop a foundation model for discovering cancer imaging biomarkers.
- To evaluate the model's performance in clinical applications.
- To assess the model's efficiency and stability compared to conventional methods.
Main Methods:
- Trained a convolutional encoder using self-supervised learning on 11,467 radiographic lesions.
- Evaluated the foundation model on distinct cancer imaging biomarker discovery tasks.
- Compared performance against supervised and state-of-the-art pretrained models.
Main Results:
- The foundation model facilitated efficient learning of imaging biomarkers.
- Task-specific models significantly outperformed conventional methods, especially with limited data.
- The model demonstrated enhanced stability and strong biological associations.
Conclusions:
- Foundation models hold significant potential for novel imaging biomarker discovery in oncology.
- This approach can accelerate the clinical translation of imaging biomarkers.
- The methodology may be applicable to other clinical use cases beyond cancer imaging.
