Implementation of An Automated Workflow for Planning Tangential Breast Radiotherapy
AAPM ePoster Library. Keiper T. 07/25/21; 323372; 58532
Poster accessible from AAPM Meeting App Only
RESERVED CONTENT
RESERVED CONTENT
This content is exclusively accessible to registered users.
Abstract
T Keiper*, D Hoffman, K Kisling, UC San Diego, La Jolla, CA
Purpose: Automated planning for breast radiotherapy can improve efficiency and consistency for one of the most commonly treated disease sites. While several in-house solutions have been proposed, few have been translated to other clinical practices. This work aims to implement a previously developed auto-planning workflow for tangential breast radiation in an independent clinic.
Methods: Each step of planning is automated, including the beam geometry. The contours are generated via atlas-based auto-segmentation in MIM. The posterior border of the tangents and thus the beam geometry is determined with a support vector machine in MATLAB. Eclipse scripts create the plan in the TPS, and EZFluence optimizes the dose distribution. Breast plans were generated for 12 patients, 4 to tune the beam generation parameters and 8 for independent testing. Auto-plan quality was compared to the clinical plan using DVH metrics. Beam geometry similarity and treatment coverage were assessed with the 50% and 90% isodose overlap, respectively, using the Dice similarity coefficients (DSC) and mean absolute distance (MAD).
Results: All DVH parameters were comparable to the clinical plans. On average, the heart mean dose changed by 0±13cGy, the lung V20Gy by +1.4±3.8%, and D95% of the tumor bed by +1.0±1.2% of prescription. The average maximum dose was +0.5±1.0% and the homogeneity (V105%/V95%) was 0.02±0.04. We confirmed similarity of beam geometry and coverage with DSC averaging 0.91(50%: ±0.06, 90%: ±0.07) and MAD averaging 0.27cm(50%: ±0.23cm, 90%: ±0.21cm).
Conclusion: We successfully integrated a suite of commercial and in-house tools for breast auto-planning at our institution. The methodology for determining beam geometry, first developed at an outside institution, was implemented in our workflow, showing the generalized applicability of this approach. Automated plans were comparable to clinical plans. Next, we will implement this automated workflow in our clinic and assess the efficiency gains.
Funding Support, Disclosures, and Conflict of Interest: Funding support from AHRQ R01HS025440
Keywords
breast,radiation therapy,treatment planning
Taxonomy
TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation
Purpose: Automated planning for breast radiotherapy can improve efficiency and consistency for one of the most commonly treated disease sites. While several in-house solutions have been proposed, few have been translated to other clinical practices. This work aims to implement a previously developed auto-planning workflow for tangential breast radiation in an independent clinic.
Methods: Each step of planning is automated, including the beam geometry. The contours are generated via atlas-based auto-segmentation in MIM. The posterior border of the tangents and thus the beam geometry is determined with a support vector machine in MATLAB. Eclipse scripts create the plan in the TPS, and EZFluence optimizes the dose distribution. Breast plans were generated for 12 patients, 4 to tune the beam generation parameters and 8 for independent testing. Auto-plan quality was compared to the clinical plan using DVH metrics. Beam geometry similarity and treatment coverage were assessed with the 50% and 90% isodose overlap, respectively, using the Dice similarity coefficients (DSC) and mean absolute distance (MAD).
Results: All DVH parameters were comparable to the clinical plans. On average, the heart mean dose changed by 0±13cGy, the lung V20Gy by +1.4±3.8%, and D95% of the tumor bed by +1.0±1.2% of prescription. The average maximum dose was +0.5±1.0% and the homogeneity (V105%/V95%) was 0.02±0.04. We confirmed similarity of beam geometry and coverage with DSC averaging 0.91(50%: ±0.06, 90%: ±0.07) and MAD averaging 0.27cm(50%: ±0.23cm, 90%: ±0.21cm).
Conclusion: We successfully integrated a suite of commercial and in-house tools for breast auto-planning at our institution. The methodology for determining beam geometry, first developed at an outside institution, was implemented in our workflow, showing the generalized applicability of this approach. Automated plans were comparable to clinical plans. Next, we will implement this automated workflow in our clinic and assess the efficiency gains.
Funding Support, Disclosures, and Conflict of Interest: Funding support from AHRQ R01HS025440
Keywords
breast,radiation therapy,treatment planning
Taxonomy
TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation
T Keiper*, D Hoffman, K Kisling, UC San Diego, La Jolla, CA
Purpose: Automated planning for breast radiotherapy can improve efficiency and consistency for one of the most commonly treated disease sites. While several in-house solutions have been proposed, few have been translated to other clinical practices. This work aims to implement a previously developed auto-planning workflow for tangential breast radiation in an independent clinic.
Methods: Each step of planning is automated, including the beam geometry. The contours are generated via atlas-based auto-segmentation in MIM. The posterior border of the tangents and thus the beam geometry is determined with a support vector machine in MATLAB. Eclipse scripts create the plan in the TPS, and EZFluence optimizes the dose distribution. Breast plans were generated for 12 patients, 4 to tune the beam generation parameters and 8 for independent testing. Auto-plan quality was compared to the clinical plan using DVH metrics. Beam geometry similarity and treatment coverage were assessed with the 50% and 90% isodose overlap, respectively, using the Dice similarity coefficients (DSC) and mean absolute distance (MAD).
Results: All DVH parameters were comparable to the clinical plans. On average, the heart mean dose changed by 0±13cGy, the lung V20Gy by +1.4±3.8%, and D95% of the tumor bed by +1.0±1.2% of prescription. The average maximum dose was +0.5±1.0% and the homogeneity (V105%/V95%) was 0.02±0.04. We confirmed similarity of beam geometry and coverage with DSC averaging 0.91(50%: ±0.06, 90%: ±0.07) and MAD averaging 0.27cm(50%: ±0.23cm, 90%: ±0.21cm).
Conclusion: We successfully integrated a suite of commercial and in-house tools for breast auto-planning at our institution. The methodology for determining beam geometry, first developed at an outside institution, was implemented in our workflow, showing the generalized applicability of this approach. Automated plans were comparable to clinical plans. Next, we will implement this automated workflow in our clinic and assess the efficiency gains.
Funding Support, Disclosures, and Conflict of Interest: Funding support from AHRQ R01HS025440
Keywords
breast,radiation therapy,treatment planning
Taxonomy
TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation
Purpose: Automated planning for breast radiotherapy can improve efficiency and consistency for one of the most commonly treated disease sites. While several in-house solutions have been proposed, few have been translated to other clinical practices. This work aims to implement a previously developed auto-planning workflow for tangential breast radiation in an independent clinic.
Methods: Each step of planning is automated, including the beam geometry. The contours are generated via atlas-based auto-segmentation in MIM. The posterior border of the tangents and thus the beam geometry is determined with a support vector machine in MATLAB. Eclipse scripts create the plan in the TPS, and EZFluence optimizes the dose distribution. Breast plans were generated for 12 patients, 4 to tune the beam generation parameters and 8 for independent testing. Auto-plan quality was compared to the clinical plan using DVH metrics. Beam geometry similarity and treatment coverage were assessed with the 50% and 90% isodose overlap, respectively, using the Dice similarity coefficients (DSC) and mean absolute distance (MAD).
Results: All DVH parameters were comparable to the clinical plans. On average, the heart mean dose changed by 0±13cGy, the lung V20Gy by +1.4±3.8%, and D95% of the tumor bed by +1.0±1.2% of prescription. The average maximum dose was +0.5±1.0% and the homogeneity (V105%/V95%) was 0.02±0.04. We confirmed similarity of beam geometry and coverage with DSC averaging 0.91(50%: ±0.06, 90%: ±0.07) and MAD averaging 0.27cm(50%: ±0.23cm, 90%: ±0.21cm).
Conclusion: We successfully integrated a suite of commercial and in-house tools for breast auto-planning at our institution. The methodology for determining beam geometry, first developed at an outside institution, was implemented in our workflow, showing the generalized applicability of this approach. Automated plans were comparable to clinical plans. Next, we will implement this automated workflow in our clinic and assess the efficiency gains.
Funding Support, Disclosures, and Conflict of Interest: Funding support from AHRQ R01HS025440
Keywords
breast,radiation therapy,treatment planning
Taxonomy
TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation
Code of conduct/disclaimer available in General Terms & Conditions
{{ help_message }}
{{filter}}