Bartłomiej W. Papież

Segmentation models implicitly learn powerful object-centric representations that outperform encoder-based feature extraction for medical image prediction, even without explicit downstream supervision.

Despite the promise of multimodal fusion, existing architectures struggle to surpass single-tracer performance in PET/CT segmentation, particularly for FDG.

Generated retinal images can inherit critical clinical information, but they may not perform well against real-world classifiers, exposing a crucial representation gap.

Jointly predicting mutations in lung cancer can significantly boost diagnostic accuracy, but only for the right combinations of genes.