RT-SRTS: Angle-agnostic real-time simultaneous 3D reconstruction and tumor segmentation from single X-ray projection

Area of Science:

• Medical Imaging
• Radiotherapy
• Artificial Intelligence

Background:

• Respiration-induced organ motion hinders radiotherapy accuracy.
• Current 3D imaging methods lack direct tumor localization and fixed-angle validation.
• Real-time motion control in radiotherapy requires advanced imaging solutions.

Purpose of the Study:

• To develop a novel imaging method for simultaneous 3D reconstruction and tumor segmentation.
• To address limitations of existing methods in tumor localization and angle variability.
• To enable real-time tumor tracking for enhanced radiotherapy.

Main Methods:

• Proposed RT-SRTS method integrating 3D imaging and tumor segmentation using multi-task learning (MTL).
• Introduced Attention Enhanced Calibrator (AEC) and Uncertain-Region Elaboration (URE) modules for improved feature extraction and segmentation.
• Validated the method on fifteen patient cases, comparing against state-of-the-art techniques.

Main Results:

• RT-SRTS achieved superior 3D reconstruction and accurate tumor segmentation.
• Simultaneous reconstruction and segmentation completed in approximately 70 ms, meeting real-time requirements.
• Ablation studies confirmed the effectiveness of the AEC and URE modules.

Conclusions:

• RT-SRTS offers a significant advancement in real-time 3D imaging and tumor segmentation for radiotherapy.
• The method effectively overcomes challenges posed by organ motion and imaging angle variability.
• This approach holds promise for improving the precision and efficacy of cancer treatment.