A computer-aided detection method for early lung cancer that adapts the Segment Anything Model (SAM) to lung-nodule imaging via transfer learning. Its novel contribution is driving SAM's promptable vision-transformer architecture with bounding-box prompts and fine-tuning it to the CT lung-nodule domain, producing precise nodule masks that feed a malignancy classifier -- instead of training a dedicated segmentation network from scratch. Reported performance is a Dice Similarity Coefficient of 97.08%, an Intersection-over-Union of 95.6% for segmentation, and 96.71% classification accuracy, demonstrating that a segmentation foundation model plus transfer learning can strengthen CAD systems for early lung cancer detection.
Proposes a computer-aided detection method for early lung cancer diagnosis that segments and classifies lung nodules in CT scans by adapting the Segment Anything Model (SAM) with transfer learning. R…