(Thursday, 8/2/2018) 7:30 AM - 9:30 AM
Room: Davidson Ballroom B
Purpose: Current methods of characterizing dose distributions assume physiologic homogeneity within a region of interest (ROI) and do not provide a spatially aware description of dose. The practice of radio-morphology (RM) is proposed as a method to apply anatomical knowledge to parametrically derive structures from a normalized set of anatomy, to produce consistently identifiable features. The goal of this study was to design a method to identify characteristics of a dose distribution in anatomical sub-volumes that are predictive of post-treatment outcomes.
Methods: The RM pipeline consists of three steps: anatomy normalization, image transformation, and dose feature extraction. First, ROIs are deformed to a standard coordinate frame, allowing them to be compared across a patient population. Next, scaling and partitioning transformations can be applied to derive new structures. Finally, radiation dose is mapped onto the structures to extract dose statistics. Using data from a learning health system, these methods were used to explore the spatial effects of dose to loss and recovery of salivary function using parotid and submandibular glands and the incidence of impotence using dose to neuro-vasculature around the prostate.
Results: The RM pipeline produced feature sets that uncovered the spatial importance of dose to the anatomy. High dose to the superior-anterior region of the contralateral parotid gland was found to promote the development of high-grade post-treatment xerostomia. High dose to the neuro-vasculature inferior to the prostate (lower-urethral-sphincter) was linked to loss of sexual function.
Conclusion: The RM pipeline is used to model the effect of radiation dose on post-treatment clinical outcomes. By parameterizing feature generation, it is possible to iteratively optimize features and methodically identify key anatomic regions. With modern databases storing full dose distributions and patient assessments, the proposed feature generation pipeline can rapidly lead to new insight about the quality and toxicity risks of radiotherapy plans.
Funding Support, Disclosures, and Conflict of Interest: Partial support provided by the Radiation Oncology Institute. T McNutt acknowledges funding support from Philips and Toshiba Medical Systems Corp. J Lee acknowledges funding support from Toshiba Medical Systems Corp and Johns Hopkins Radiation Oncology Discovery Award.
Keywords
Feature Extraction, Image Processing
Taxonomy
TH- response assessment : Machine learning
Contact Email
