Fig. 4: Saliency map of the breast age prediction model.

4-view (bilateral CC and MLO) mammograms from the inhouse dataset are shown. Saliency maps highlight regions that the model focuses on when predicting breast age. Age-specific saliency maps were computed by categorizing the data into five sub-age groups, delineated by 10-year intervals, ranging from ≤40 to ≥70 years, and into four density groups (A–D) based on ACR density classification for mammograms. Within each subgroup, different-sized masks (32 × 32, 64 × 64, 128 × 128, and 256 × 256) were used for the analysis. Normalization was carried out by dividing the entire image by the maximum value, ensuring that the values of the final saliency map ranged from 0 to 1. The minimum value within this range corresponds to blue color on the colorimetric map, while the maximum value corresponds to red. The saliency maps reveal that the model commonly focuses on features such as breast skin thickness, fibroglandular tissue, calcifications, masses, and breast vessels. The MLO views generally provide more informative features related to aging compared to the CC views.