Advancing Radiotherapy Treatment Through Artificial Intelligence-Driven Approaches

Radiotherapy is a critical treatment regimen for cancer patients. Modern radiotherapy aims to deliver precise radiation doses to specific tumor target volumes while minimizing exposure to surrounding healthy tissues and organs. To achieve precision and high treatment quality, accurate target delineation, streamlined workflows, and timely follow-up are necessary. However, the current clinical standard of radiotherapy has limitations that need to be addressed. These include the slow manual delineation of targets and organs-at-risk, a lack of tools for early prediction of treatment outcomes, and the potential radiation toxicity and side effects. Fortunately, ultra-high dose rates (FLASH) irradiation emerges as a promising new modality of radiotherapy that has the potential to reduce the radiation toxicity to surrounding normal tissues while maintaining tumor control. However, the clinical translation of FLASH is challenging, and precise quantification of radiobiology is urgently needed. To address these above limitations and needs, this dissertation aims to develop artificial intelligence (AI)-driven techniques to advance current radiotherapy treatment. Recent breakthroughs in AI, including mathematical algorithms and high-performance computing technologies, have led to transformative impacts in healthcare. Various studies have shown significant improvements in the quality and efficiency of image segmentation and treatment outcome analysis with AI approaches. Therefore, this dissertation specifically focuses on leveraging the power of AI to improve the target delineation, predict treatment outcomes, and assist the clinical translation of FLASH in the following three parts: (1) Streamlining and standardizing Stereotactic Radiosurgery (SRS) workflow with AI. In this part, an AI-driven auto-segmentation and labeling platform was firstly developed for SRS patients with multiple brain metastases (mBMs). This platform can automatically segment out mBMs and auto-label each segmentation with an atlas label in high accuracy compared with manual contours and labels. Secondly, a deep-learning and radiomics ensemble classifier was developed for the false positive reduction in the raw mBMs segmentation, to improve the mBMs detection specificity while maintaining a promising sensitivity. (2) Effective SRS management through AI predictions. In this part, three AI techniques for SRS treatment outcome prediction were developed including an ensemble learning model for glioma patients overall survival prediction, an unsupervised structure learning model for mBMs SRS treatment response modeling, and an AI model for mBMs patients post-SRS neurocognitive decline prediction, to assist the SRS treatment decision-making and to improve the treatment quality. (3) Quantitative FLASH radiotherapy with AI. In this part, an AI model was developed to estimate the equivalent conventional (CONV) dose of FLASH irradiation based on tissue histological images. This developed model can accurately estimate the equivalent dose in CONV irradiation and indicates that deep learning can be potentially used to assess the equivalent dose of FLASH irradiation to normal tissue to accelerate its clinical implementation. In summary, this dissertation presents novel AI technique developments that could potentially address the current needs in radiotherapy treatment. Moreover, the proposed approaches can be transferred to different tumor sites and radiotherapy procedures, thus generating broader clinical impact.

Author dc:creator

• Yang, Zi

Contributors dc:contributor

• Lu, Weiguo
• Gu, Xuejun
• Wang, Jing
• Jia, Xun
• Wardak, Zabi

• Brain Neoplasms
• Outcome Assessment, Health Care
• Radiosurgery
• Radiotherapy
• Artificial Intelligence

Language dc:language

en

Identifier

1522122394

OAI identifier oai:identifier

oai:utswmed-ir.tdl.org:2152.5/10584