Click here to

Session: Multi-Disciplinary General ePoster Viewing [Return to Session]

A Knowledge-Based Automatic Lung IMRT Planning Method for Partial Heart Sparing

L Yuan*, J Sohn, R Singh, E Weiss, S Kim, Virginia Commonwealth University, Richmond, VA


PO-GePV-M-233 (Sunday, 7/25/2021)   [Eastern Time (GMT-4)]

Purpose: Recent findings suggest that sparing of cardiac sub-volumes might be critical to maintain normal organ function. However, a lack of established dose constraints and planning experience hinder the clinical adaptation of cardiac substructure sparing in IMRT planning. This study was to establish clinically achievable cardiac substructure dose constraints based on prior plans and investigate their correlation with patient anatomy. An automatic lung IMRT planning method was developed to incorporate patient specific normal tissue dose constraints.

Methods: Seventeen locally-advanced lung cancer patients who were treated with fixed-field IMRT were included in this study. The cardiac-substructures and the right heart sub-volume were contoured. Knowledge models were trained to predict the OARs and cardiac-substructure dose metrics utilized a step-wise regression method to quickly build models with good generalizability using only limited dataset. The predicted dose metrics were compared with the actual dose in clinical plans and a leave-one-out cross-validation method was used to evaluate the prediction accuracy of the dose metrics. The optimization objectives for automatic planning were determined based on model prediction. The creation of the objectives and plan optimization were all handled by a computer script interfacing with the treatment planning system with no human intervention. The dose metrics obtained by the automatic plans are compared with the clinical plans by paired signed-rank tests.

Results: Knowledge models showed good predictive accuracy: the determination coefficients (R2) for the dose metrics range between 0.3-0.9. Eight of the twelve predicted dose metrics had R2>0.5. In particular, R2 was 0.81 for right sub-volume of heart. Automatic plans improved all heart dose metrics and esophagus V60Gy and were comparable for all other dose metrics.

Conclusion: Knowledge models can be utilized to determine patient-specific dose constraints for heart sub-volume and other OARs. Automatic planning can achieve comparable or better plan quality than clinical plans.

Funding Support, Disclosures, and Conflict of Interest: E. Weiss: NIH research grant, UpToDate royalties, Viewray research funding



    Heart, Lung


    TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation

    Contact Email