Systems and methods for accelerated online adaptive radiation therapy
Inventors
LI, X. ALLEN • Zhang, Ying • Lim, Sara • Zhang, Jingqiao • Ahunbay, Ergun • Thapa, Ranjeeta • Nasief, Haidy
Assignees
Publication Number
US-12005270-B2
Publication Date
2024-06-11
Expiration Date
2039-06-26
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
Systems and methods for accelerated online adaptive radiation therapy (“ART”) are described. The improvements to online ART are generally provided based on the use of textural analysis and machine learning algorithms implemented with a hardware processor and a memory. The described systems and methods enable more efficient and accurate online adaptive replanning (“OLAR”), which can also be implemented in clinically acceptable timeframes. For example, OLAR can be reduced from taking 10-30 minutes down to 5-10 minutes.
Core Innovation
The invention provides systems and methods for validating and correcting the accuracy of radiation treatment plan contour data using automated image-texture analysis and machine learning. This approach involves accessing existing contour data associated with a radiation treatment plan and the corresponding subject image, generating inner and outer shell regions by eroding and expanding the contours, and extracting image feature data from these regions. The image feature maps of both shells are evaluated using machine learning algorithms, such as decision tree or recursive random forest classification, to validate whether the contours are accurate or inaccurate.
Further, the patent addresses a significant problem in online adaptive radiation therapy—namely, the need for efficient and robust validation of automated contour segmentations, which traditionally relied on time-consuming manual reviews and led to impractical treatment delays. By comparing texture-based image features in the shells and core regions of the contours against ground truth data, the system automatically labels contours as accurate or inaccurate and, if necessary, triggers a correction procedure that applies texture-based active contour algorithms to improve contour accuracy.
This automated validation and correction method supports the clinical workflow by enabling more rapid, consistent, and objective assessment of anatomical contours for radiation therapy planning. The process is implemented using computer systems and can be integrated with existing medical image sources and radiation treatment planning systems, allowing for a streamlined and accelerated adaptive replanning process.
Claims Coverage
The independent claims describe four main inventive features directed at validating and correcting radiation treatment plan contour data using shell-based image texture analysis and machine learning.
Automated contour validation using inner and outer shell texture features
This feature enables a computer system to: - Access pre-existing contour data and associated subject images. - Generate inner shell data by eroding contours by a first margin and outer shell data by expanding contours by a second margin. - Compute image feature data (including intensity histogram and texture features) from the subject image. - Create inner and outer shell image feature maps and input these to a machine learning algorithm, which outputs labeled contour data indicating whether the contour is validated as accurate or inaccurate. - This process includes generating a binary mask for each contour, and creating core region data by determining pixels within the contour with the greatest distance from the contour boundary.
Contour validation with same-value margin for inner and outer shells
This feature specifies generating both inner and outer shells with the same margin value (e.g., 4 mm) for efficient and consistent validation of contours. The image feature maps from these shells are input to a machine learning algorithm to label contours as accurate or inaccurate.
Recursive random forest-based image feature selection for contour validation
This feature involves using a machine learning algorithm implementing a recursive random forest classification to select a set of image features that best enable the comparison of inner and outer shell image feature maps for accurate validation of contour data.
Automated contour correction for inaccurate contours with re-validation
When contours are labeled as inaccurate, the system uses a contour correction algorithm that: - Computes second image feature data for a region-of-interest defined by the inaccurate contour. - Generates an image feature map for the ROI. - Utilizes the image feature map as external force in an active contour algorithm to adjust and correct the contour boundary. - Following correction, the corrected contour data are re-validated using the same shell-based machine learning approach.
The claims collectively protect a suite of shell-based, image-feature-driven, and machine learning-enabled methods for automated, accurate contour validation and correction in radiation therapy planning.
Stated Advantages
Enables significantly faster and automated validation and correction of contour data, reducing reliance on time-consuming manual review and editing.
Facilitates objective and patient-specific accuracy assessment of radiation treatment contours using quantitative texture analysis.
Improves clinical efficiency by accelerating the online adaptive replanning workflow within clinically acceptable timeframes.
Documented Applications
Automated validation and correction of anatomical contour data in radiation treatment planning for online adaptive radiation therapy (ART) and online adaptive replanning (OLAR).
Interested in licensing this patent?