In press
Foundation-Model Imaging Biomarkers for Pancreatic Cancer: Cross-Species Discovery of Conserved Tumour Biology in Mouse MRI and Human CT
Yu G, et al.
eBioMedicine
Frozen foundation-model embeddings recover matched biological endpoints from both mouse 9.4T MRI and human clinical CT, outperforming hand-crafted radiomics, end-to-end fine-tuning, and a purpose-built 3D-CT vision-language model.
Under review
A Systematic Review of Reward Design and Evaluation in Reinforcement-Learning Post-Training of Medical Imaging Models
Yu G, et al.
npj Digital Medicine
2025
Yu G, Zhang Z, Eresen A, Hou Q, Yaghmai V, Zhang Z.
Diagnostics 15(22):2844
U-Net++ with an EfficientNet-B0 encoder segments the tumor and predicts response in one pass, reaching Dice 0.92 and AUROC 0.97, with the imaging biomarkers validated against H&E viability and TUNEL apoptosis.
Code
AI-Derived MRI Biomarkers Using Vision Transformer for Predicting Combination Immunotherapy Outcomes in Liver Cancer
Yu G, et al.
ISMRM & ISMRT Annual Meeting
Zhang Z, Yu G, Hou Q, et al.
Cancers 17(15):2477
Zhang Z, Yu G, et al.
Cancers 17(4):699
2024
Yu G, et al.
Journal of Translational Medicine 22(1):76
743 radiomic features reduced to five. An SVM on T1w and T2w reaches AUC 1.00 for binary response and 0.93 across outcome classes, with MRI-derived measures tracking histology at r = 0.94.
Yu G, Zhang Z, Eresen A, et al.
International Journal of Molecular Sciences 25(22):12038
Eresen A, Zhang Z, Yu G, et al.
BMC Cancer 24:1215
Eresen A, Zhang Z, Yu G, et al.
American Journal of Cancer Research 14(1):344–354
Zhang Z, Yu G, et al.
American Journal of Cancer Research 14(5):2216–2227
Zhang Z, Yu G, et al.
Annals of Translational Medicine 12(4):77
2022
Gan W, Lin Y, Yu G, et al. Co-first author
Proceedings of SemEval-2022, ACL
An XLM-RoBERTa system with continued pretraining, adversarial training, and ensembling. Third in the multilingual track, fourth in code-mixed, seventh in Chinese.