Lung cancer remains one of the most challenging cancers to treat and continues to be a leading cause of cancer-related mortality worldwide. Innovations in radiation oncology are critical to enhancing treatment precision and improving patient outcomes.
This PhD project — a collaboration between HEGP / APHP Hospital, Université Paris Cité, and TheraPanacea — aims to leverage artificial intelligence (AI) to tackle key challenges in lung cancer management. The research is structured into three interconnected work packages, each addressing specific aspects of treatment planning, delivery, and assessment.
Objective 1: AI-Driven Target Volume Delineation
The first objective focuses on developing a deep learning method for automatic delineation of target volumes in CT images. Accurate segmentation of tumors and surrounding organs at risk is essential for precise radiation therapy. However, manual contouring is time-consuming, labor-intensive, and prone to inter-observer variability. This work package will develop a convolutional neural network (CNN) optimized for 3D medical imaging to segment lung tumors and critical structures with high accuracy. Where available, multi-modal imaging data will be incorporated to enhance segmentation precision and consistency, streamlining treatment planning and reducing variability.
Objective 2: Real-Time Tumor Motion Tracking
The second objective addresses tumor motion during radiation delivery, primarily caused by physiological factors such as respiration. A recursive neural network (RNN) will be developed to automatically track tumor movement in real time using 2D bi-planar X-rays. By leveraging the sequential nature of motion data, the model will continuously monitor tumor positioning during treatment. This innovation aims to improve tracking accuracy and incorporate respiratory motion patterns, enabling more precise radiation delivery while minimizing exposure to healthy tissue.
Objective 3: Dose Verification via 3D Reconstruction and Deep Learning-Based Dose Simulation
The third objective addresses the critical need for accurate dose verification during lung cancer radiotherapy. Traditional bi-planar imaging techniques provide limited volumetric information, posing challenges for precise assessment of delivered dose distributions. A generative deep learning model will be developed to perform high-fidelity 3D reconstruction, enabling accurate localization of tumor and organs-at-risk positions in near real time. Leveraging this reconstructed geometry, a deep learning-based dose simulation framework will then estimate the delivered dose distribution, incorporating tumor motion and anatomical changes. This approach will allow clinicians to perform adaptive dose verification, comparing planned versus delivered dose maps to assess treatment accuracy and detect potential deviations
