OncoReg Challenge's Hidden Twist: How Privacy-Preserving Image Registration Is Unlocking Better Radiotherapy

The OncoReg Challenge addresses a critical bottleneck in cancer research: underutilization of vast medical data due to patient privacy concerns. Building on the Learn2Reg Challenge, it introduces a two-phase framework for developing and validating image registration methods. Phase one uses publicly available datasets for initial model development. Phase two moves training to a private dataset within secure hospital networks, enabling researchers to create more generalizable AI models without compromising patient privacy. The challenge specifically targets registration of interventional cone-beam computed tomography (CBCT) with standard planning fan-beam CT (FBCT) in radiotherapy—a task essential for dynamic treatment adjustments in image-guided radiotherapy. Accurate alignment minimizes radiation exposure to healthy tissues while effectively targeting tumors.

A comprehensive analysis of competition entries reveals that feature extraction plays a pivotal role in this registration task. A new method demonstrated its versatility, while established approaches continued to perform comparably to newer techniques. Interestingly, both deep learning and classical approaches still play significant roles in image registration. The most effective solutions combined methods, particularly in feature extraction, suggesting that hybrid strategies are currently optimal. The study provides detailed methodology, data descriptions, and thorough evaluation of entries, offering a roadmap for future work in privacy-preserving medical image registration for oncology.