Deep learning (DL) techniques have gained substantial interest across numerous scientific fields, with healthcare emerging as a prominent area of successful application in addressing complex medical challenges. Our group is dedicated to developing innovative DL methods that emulate human-like decision-making in medical imaging and radiotherapy.
In medical imaging, our work spans image reconstruction, quality enhancement, and segmentation. We have successfully established a deep reinforcement learning framework that intelligently adjusts regularization parameters in iterative reconstruction for CT in a human-like manner. Additionally, we developed a high-quality low-dose CT reconstruction algorithm that leverages a manifold prior learned through a deep network to enhance the regularization process.
In radiotherapy, we have pioneered an Intelligent Automatic Treatment Planning framework, featuring virtual treatment planners and physicians that mimic human planning behaviors. This virtual team autonomously operates real treatment planning systems to produce clinically acceptable, high-quality treatment plans. Notably, the plan quality generated by our framework has outperformed human-generated plans in dosimetry planning challenges.
Funding Support:
- “Human-like automated radiotherapy treatment planning via imitation learning.” NIH/NCI R01CA254377, 05/2021 – 04/2026.
- “Intelligent treatment planning for cancer radiotherapy.” NIH/NCI R01CA237269, 04/2019 – 03/2024.
