Investigated Meta Pseudo Labels for breast cancer CAD. The pipeline failed because the baseline segmentation model could not learn from the extremely limited labeled data. This research demonstrates that it's success is contingent on first establishing a functional supervised model, a step that was ultimately hampered by severe data scarcity
python computer-vision deep-learning tensorflow medical-imaging semi-supervised-learning research-project image-segmentation breast-cancer meta-learning u-net efficientnet mammography cad-system data-scarcity meta-pseudo-labels model-failure-analysis
-
Updated
Jul 6, 2025 - Jupyter Notebook