A foundation model for cancer imaging biomarkers was developed by training a convolutional encoder using self-supervised learning on a comprehensive dataset of 11,467 radiographic lesions. The model was evaluated in three distinct clinical applications: lesion anatomical site classification, nodule malignancy prediction, and NSCLC prognosis. The foundation model outperformed conventional supervised and state-of-the-art pretrained models, especially in scenarios with limited training data. It showed strong stability to input variations and strong associations with underlying biology. The model was implemented using two approaches: (1) training a linear classifier on extracted features and (2) transfer learning by fine-tuning all model parameters. The foundation model demonstrated superior performance in all three use cases, with significant improvements in balanced accuracy and mAP compared to baseline methods. The model's features were also found to be highly stable against inter-reader and test–retest variations. Additionally, the model's features showed strong associations with immune-related biological pathways. The study highlights the potential of foundation models in discovering new imaging biomarkers and their application in various clinical settings. The foundation model was implemented using a modified version of the SimCLR framework, with modifications to the transformation, encoder, and loss functions to suit medical imaging. The model was pretrained for 100 epochs using a batch size of 64 on two NVIDIA GPUs. The foundation model was then adapted for specific tasks through feature extraction and transfer learning. The model's performance was evaluated on three distinct use cases, with significant improvements in balanced accuracy and mAP compared to baseline methods. The model's features were also found to be highly stable against inter-reader and test–retest variations. The study demonstrates the potential of foundation models in discovering new imaging biomarkers and their application in various clinical settings.A foundation model for cancer imaging biomarkers was developed by training a convolutional encoder using self-supervised learning on a comprehensive dataset of 11,467 radiographic lesions. The model was evaluated in three distinct clinical applications: lesion anatomical site classification, nodule malignancy prediction, and NSCLC prognosis. The foundation model outperformed conventional supervised and state-of-the-art pretrained models, especially in scenarios with limited training data. It showed strong stability to input variations and strong associations with underlying biology. The model was implemented using two approaches: (1) training a linear classifier on extracted features and (2) transfer learning by fine-tuning all model parameters. The foundation model demonstrated superior performance in all three use cases, with significant improvements in balanced accuracy and mAP compared to baseline methods. The model's features were also found to be highly stable against inter-reader and test–retest variations. Additionally, the model's features showed strong associations with immune-related biological pathways. The study highlights the potential of foundation models in discovering new imaging biomarkers and their application in various clinical settings. The foundation model was implemented using a modified version of the SimCLR framework, with modifications to the transformation, encoder, and loss functions to suit medical imaging. The model was pretrained for 100 epochs using a batch size of 64 on two NVIDIA GPUs. The foundation model was then adapted for specific tasks through feature extraction and transfer learning. The model's performance was evaluated on three distinct use cases, with significant improvements in balanced accuracy and mAP compared to baseline methods. The model's features were also found to be highly stable against inter-reader and test–retest variations. The study demonstrates the potential of foundation models in discovering new imaging biomarkers and their application in various clinical settings.