Abstract:
Aim: This study delves into cutting-edge cancer treatments by leveraging advancements in quantum biology, genomics, and artificial intelligence. Additionally, it investigates the integration of AI- generated auto-contours, using deep learning segmentation, into diverse treatment planning systems.
Introduction: Despite advancements in cancer treatment—from deep X-ray therapies to high-energy photons, electrons, protons, carbon ions, and neutrons—the cure rate remains a significant challenge. The core issue lies within the biological domain, which requires more focused attention. Genomics holds the key to discovering robust solutions for cancer treatment, with the integration of physics into genomics being a pivotal next-generation approach. The combination of AI and physics offers the most promising outcomes in genomics. Quantum biology, which applies quantum mechanical principles to biological systems, could revolutionize cancer therapy in the future by addressing many of its unresolved challenges. Until such breakthroughs are realized, it is crucial to maximize the effectiveness of current technologies. In this context, artificial intelligence and machine learning should be fully leveraged. In our study, we have applied AI-generated auto-contours, utilizing deep learning segmentation, across diverse treatment planning systems. This approach enhances treatment outcomes, improves accessibility to remote areas, and provides financial benefits.
Material and Methods: The study utilized the Ray Station planning system 12A (Ray Search Laboratories, Sweden), renowned for its GPU-powered algorithm capable of generating AI-generated contours through deep learning segmentation. The research encompassed a group of hospitals comprising five facilities equipped with Eclipse V16.1 (Varian Medical Systems, USA) and Monaco V6.1.2 (Elekta Medical Systems, Crawley, UK) treatment planning systems, distributed across various locations in India. Additionally, a central planning system utilizing Ray Station TPS was deployed at a distinct location. Simulated CT images for radiation oncology (RO) planning were transmitted to the cloud and subsequently imported into the Ray Station platform. Auto contours were then generated on these CT
images and exported back to the respective TPS via cloud connectivity. Importantly, this process enabled the seamless transfer of auto contoured images from the cloud to both Eclipse and Monaco contour stations, ensuring consistency and interoperability across diverse treatment planning environments. The study analyzed over 500 cases across these five units, encompassing various diagnoses, to assess the efficacy of this approach.
Results: The OAR contours generated through deep learning segmentation in Ray Station were seamlessly transferred to both Monaco and Eclipse TPS via cloud connectivity. Analysis revealed that 98% of the contours were deemed perfect and utilized in clinical planning. The remaining 2% of errors were primarily attributed to factors such as patient movement and image clarity. Notably, the average time required for each auto contour was less than 2 minutes.
Conclusions: The Ray Station planning system, leveraging a GPU-powered algorithm and deep learning segmentation, proves instrumental in generating AI-generated auto contours for all organs at risk (OARs). Importantly, this technology demonstrates its versatility by seamlessly integrating into other planning systems. The efficiency gains realized through this tool not only translate to significant time savings but also ensure uniformity of contours across all our units. This consistency fosters enhanced quality in treatment planning, facilitates research endeavour, and ultimately contributes to improved patient care especially in developing countries where the budget for dedicated treatment planning systems are not adequate.
Key words: AI, Auto Contours, Ray Station, Treatment Planning Systems (TPS), Deep Learning Segmentation, Cloud Connectivity
