On November 11, 2023, the prestigious IEEE Nuclear Science Symposium and Medical Imaging Conference (IEEE NSS/MIC) was held in Vancouver, Canada. This event witnessed a significant achievement for the Shanghai Jiao Tong University's School of Biomedical Engineering. The team led by Professor Huang Qiu, with standout student Chen Gaoyu, clinched the championship at the 2nd Ultra-Low Dose PET Imaging Challenge (UDPET).

This victory highlights the profound application of Artificial Intelligence (AI) in the field of nuclear medicine. Nuclear medicine imaging, known for its higher noise levels compared to radiographic techniques, poses unique challenges. This is particularly true for special patient groups needing frequent PET scans, sensitive to radiation, or unable to maintain a fixed position, where low-dose or rapid acquisition techniques are advantageous. The UDPET Challenge, organized in collaboration with Ruijin Hospital affiliated with Shanghai Jiao Tong University Medical School and the Inselspital of the University of Bern, Switzerland, provided 1,447 paired high and low-dose PET datasets from United Imaging Healthcare’s uExplorer and Siemens' Biograph Vision Quadra systems. The low-dose images had varied down-sampling ratios, ranging from 4x to 100x.

This year's challenge upped the ante by including test datasets with unspecified doses and mismatched training data, better simulating real-world clinical scenarios and demanding higher model robustness and generalization. To tackle this, Professor Huang's team, building on their extensive experience in PET imaging and lessons from the first challenge, developed a novel Block-based Noise Level Estimation model (NLE-Net). By effectively utilizing the diverse dosage data provided, they trained their model to adapt across different doses, leading to their remarkable victory.