Cloud Dose Prediction Model Aggregation for Radiotherapy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Small clinics face challenges in creating radiotherapy treatment plans due to lack of access to training data and communication links, making automatic planning difficult, and manual optimization is time-consuming and inefficient, especially when predicting radiation dosage and dose distribution for diverse patient geometries.
Innovation Solution
A method for automatically creating a dose prediction model by aggregating clinical knowledge from multiple sources without direct communication links, using a cloud-based system where clinics submit anonymous or condensed patient data to generate a prediction model, allowing for automatic treatment plan generation without manual configuration or training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If existing systems use manual training of algorithms by human operators, then automatic planning can be achieved, but it requires access to existing radiotherapy treatment plans which is prohibitive for beginning clinics
Solution Approach 1:
The patent merges training data from multiple clinics into a centralized database, allowing individual clinics to benefit from aggregated data without each clinic needing to maintain its own extensive training dataset. This combines resources across organizational boundaries to solve the data scarcity problem for beginning clinics.
Solution Approach 2:
The patent introduces a centralized server as an intermediary that receives, stores, and distributes training data and model information between multiple clinics. This intermediary eliminates the need for direct communication links between all participants while still enabling collaborative model training and sharing.
2Reliability
If treatment plans are manually optimized by human operators, then clinical acceptability can be achieved, but the process is time consuming and inefficient
Solution Approach 1:
The system performs preliminary dose prediction using trained models before final treatment plan generation. This preliminary action provides an initial estimate of dose distribution that guides subsequent optimization, reducing the iterative cycles needed to achieve clinical acceptability.
Solution Approach 2:
The patent implements feedback loops where predicted dose distributions are compared against clinical constraints and objectives, and this information feeds back into the optimization process. This automated feedback mechanism accelerates the optimization process while maintaining clinical quality standards.
3Adaptability or versatility
If clinics share individual dose prediction models, then model availability increases, but it results in tens or hundreds of overlapping models making clinical use tedious and inefficient
Solution Approach 1:
The patent merges multiple individual clinic models into a single centralized model trained on aggregated data from all participating clinics. This consolidation maintains the benefits of diverse training data while eliminating the complexity of managing and selecting among numerous individual models.
Solution Approach 2:
The centralized model serves multiple clinics simultaneously with a single unified model, making the system universal. This multi-functional approach allows one model to address the needs of various clinics with different data characteristics, eliminating the need for each clinic to maintain separate models.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention proposes a method for automatically creating a dose prediction model based on existing clinical knowledge that is accumulated from multiple sources without collaborators establishing communication links between each other. According to embodiments of the claimed subject matter, clinics can collaborate in creating a dose prediction model by submitting their treatment plans into a remote computer system (such as a cloud-based system) which aggregates information from various collaborators and produces a model that captures clinical information from all submitted treatment plans. According to further embodiments, the method may contain a step where all patient data submitted by a clinic is made anonymous or the relevant parameters are extracted and condensed prior to submitting them over the communications link in order to comply with local regulations.