New AI Foundation Model for Disease Detection in Head CT Scans
A new foundation model called FM-CT has been created to enhance disease detection in head computed tomography (CT) imaging. This imaging technique is commonly utilized as the primary method in neurological emergencies due to its rapidity, safety, affordability, and widespread availability. FM-CT tackles the issue of limited high-quality labels and annotations for rare conditions, which has previously restricted the advancement of effective deep learning models. Utilizing self-supervised learning, the model was pre-trained on a substantial and varied dataset of 361,663 non-contrast 3D head CT scans, thus removing the necessity for manual labeling. This method equips the model to acquire strong, generalizable features. The findings, crucial for identifying various brain, skull, and cerebrovascular diseases, were shared on arXiv with the identifier arXiv:2502.02779v3, under the announcement type replace-cross.
Key facts
- FM-CT is a foundation model for generalizable disease detection in head CT scans.
- Head CT imaging is a first-line modality in neurologic emergencies due to its rapidity, safety, cost, and ubiquity.
- The model addresses the scarcity of high-quality labels and annotations for less common conditions.
- FM-CT was pre-trained using self-supervised learning on 361,663 non-contrast 3D head CT scans.
- Training did not require manual annotations, allowing the model to learn robust, generalizable features.
- Deep learning models can facilitate detection of a wide range of diseases in head CT.
- Head CT is used for assessing pathology of the brain, skull, and cerebrovascular system.
- The research was announced on arXiv with the identifier arXiv:2502.02779v3.
Entities
Institutions
- arXiv