Intelligent Decision Support for Head and Neck Radiotherapy Planning via Multi-Organ Constellation Matching

Overview

This report presents a novel knowledge-based treatment planning (KBP) infrastructure that matches current head and neck cancer radiotherapy cases to retrospective best practice cases using holistic geometric models of gross tumor volume (GTV) and organs at risk (OAR) spatial relationships. The system enables improved plan creation by providing reference plans based on similarity of multi-organ constellation geometry, addressing limitations of prior single target-OAR pair methods.

Background

Radiotherapy aims to destroy malignant tissues while sparing surrounding healthy organs at risk (OARs) to minimize adverse effects. Advances in radiotherapy delivery, such as image-guided radiotherapy and intensity-modulated radiotherapy, have increased treatment precision but made planning more complex. Knowledge-based treatment planning (KBP) tools utilize retrospective case data to optimize dose distributions, often focusing on dose-volume histogram (DVH) predictions. However, most existing KBP methods consider only single target-OAR relationships rather than holistic spatial configurations of all relevant organs.

Data Highlights

The proposed system calculates quantitative spatial features between the gross tumor volume (GTV) and all surrounding OARs to construct a relational geometric model called the GTV-OAR constellation. Similarity between cases is computed by matching these multi-organ constellation features, normalized and weighted by the number of common OARs. This approach enables identification of retrospective cases with similar anatomical geometry, which have been shown to have comparable DVHs and treatment plans.

Key Findings

• The infrastructure supports holistic modeling of spatial relationships between the GTV and multiple OARs simultaneously, unlike prior methods limited to single target-OAR pairs.
• Quantitative features capturing GTV-OAR constellation geometry are stored relationally, enabling robust comparison across cases with anatomical variation.
• Similarity matching is normalized by the number of OARs in common, allowing equitable comparison between cases.
• Cases with similar GTV-OAR constellation geometry tend to have similar dose-volume histograms, validating the approach.
• The system integrates with existing medical imaging informatics infrastructures and 3D dose simulations to facilitate clinical implementation.

Clinical Implications

This decision support infrastructure can enhance radiotherapy planning by providing clinicians with reference plans from anatomically similar cases, potentially improving plan quality and consistency. By considering the full spatial constellation of target and risk organs, the system supports more nuanced dose optimization that respects the complex anatomical context of head and neck cancers. Integration into clinical workflows may reduce planning time and variability.

Conclusion

The presented multi-organ constellation matching infrastructure represents a significant advancement in knowledge-based radiotherapy planning by enabling holistic anatomical similarity assessment. This approach holds promise to improve treatment plan quality and facilitate clinical adoption of advanced planning support tools.

References

1. Varian Medical Systems -- RapidPlan Knowledge-Based Planning
2. Literature Review on Knowledge-Based Planning (Last 15 Years)