RF01-03 Artificial intelligence-assisted screening in mammography: preliminary classification and automated prioritization Details

Start Date: 2025-09-13

End Date: 2025-09-13

Breast cancer remains a leading cause of mortality among women worldwide. Despite the impact of screening programs, mammography interpretation is time-consuming and subject to inter-reader variability, motivating decision-support tools. This study presents a proof-of-concept pipeline that retrains a convolutional neural network to classify mammograms as malignant or benign and then produces a priority ranking of all cases according to malignancy probability, enabling triage-oriented reading order. Using the MIAS dataset restricted to labeled images (n = 322), we performed metadata cleaning, class-balanced augmentation, transfer learning, model evaluation, and large-scale inference to export probabilistic scores and a ranked CSV. The model predicted 121 images as malignant and 201 as benign (threshold 0.5), with a mean malignancy probability of 0.39 ± 0.40. While performance optimization and diverse data remain necessary, these results illustrate the feasibility of an AI-assisted prioritization strategy to support clinical workflows, with future work focusing on calibration, external validation, and integration into PACS-based environments.