Oncospace

Oncospace is a software tool utilizing locked machine learning algorithms to predict patient-specific dosimetric goals and assist in configuring radiotherapy treatment plans for prostate, head, and neck cancers, integrating with existing treatment planning systems. Its methodology focuses on improving plan quality and streamlining clinical workflows by automating aspects of plan optimization, demonstrably reducing planning iterations while maintaining OAR sparing and target coverage.

Oncospace is a software device designed to assist radiation oncology professionals in configuring and reviewing radiotherapy treatment plans for patients with malignant or benign diseases in the prostate, head, and neck regions. It functions in conjunction with existing treatment planning systems (TPS), employing locked machine learning algorithms to predict patient-specific dosimetric goals for organs at risk (OARs) based on anatomical geometry. The software facilitates the setup of treatment protocols, including dose prescriptions and delivery methods, and automates the initial stages of plan optimization by transmitting these parameters and predicted OAR objectives to the TPS, while providing a user interface for plan evaluation. The underlying OAR dose prediction models have been validated using retrospective clinical data, demonstrating their capability across varied patient anatomies and treatment complexities within intensity-modulated photon beam radiotherapy for the specified regions. Performance studies indicate that Oncospace maintains target coverage and achieves non-inferior OAR dose sparing (to 5 Gy for prostate, 8 Gy for head & neck) compared to traditional planning. A significant outcome is the reported 77% average reduction in plan optimization iterations, suggesting substantial improvements in clinical workflow efficiency, although human judgment remains crucial for final plan approval and adjustments.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the performance of the machine learning model for OAR dose prediction vary with the heterogeneity of patient anatomy within the prostate, head, and neck cohorts?; What specific features within the patient-specific anatomical geometry are most heavily weighted by the machine learning algorithm when predicting dosimetric goals for OARs?; Beyond the 77% reduction in iterations, what is the quantifiable impact on the overall time from plan initiation to clinically approved plan when using Oncospace?; What is the process and data requirement for training and validating the machine learning models for new anatomical sites or significantly different treatment techniques?; How does Oncospace ensure its predicted OAR goals remain clinically relevant and do not inadvertently compromise target coverage objectives across diverse clinical scenarios?