Developed brat (Brain Report Alignment Transformer), achieving state-of-the-art performance in image-text retrieval, tumor segmentation, Alzheimer’s classification, and enabling high-quality automatic MRI report generation
Proposed RF-Deep, a random forest-based OOD detection method using deep features for robust lung cancer segmentation (FPR95 < 0.1%)
Designed a pretrained hybrid transformer for cardiac substructure segmentation in contrast/non-contrast CTs, demonstrating zero-shot generalization from lung to breast cancer patients
Demonstrated that large-scale self-supervised pretraining on heterogeneous CT scans significantly improves robustness in lung tumor segmentation (published in Medical Physics, 2025)
Introduced DAGMaN, a novel self-supervised framework combining attention-guided masked image modeling and noisy teacher co-distillation for medical imaging (Medical Imaging with Deep Learning, 2025)