MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel
  1. Home
  2. Meeting Program
  3. All
  4. An Automation Framework for Real-Time Prospective Radiomics with Deformable Image Registration

An Automation Framework for Real-Time Prospective Radiomics with Deformable Image Registration

T Patel*, P Lang , D Moseley , A Hope , D Jaffray , N Becker , T Tadic , Princess Margaret Cancer Centre, University Health Network, Toronto, ON

Presentations

(Sunday, 7/29/2018) 1:00 PM - 1:55 PM

Room: Karl Dean Ballroom C

Purpose: To develop and validate a software framework (SF) integrating several clinical systems to automate real-time radiomics analyses for patients receiving image-guided radiotherapy.

Methods: A fully-automated SF was developed to monitor patients on treatment via Mosaiq and transfer daily images (CBCT, MR) to RayStation for deformable registration with planning datasets. Patient data was processed in a Python module to compute, store and visualize radiomic features. The SF was tested by implementing a method for deriving a CBCT marker (CBCTM) previously correlated to radiation pneumonitis (RP) in NSCLC patients. CBCTM were retrospectively computed at each fraction for 129 patients and compared against a previous MATLAB implementation using Elastix and requiring manual intervention. Performance was characterized by processing time and registration success rate. Using a 20 patient subset, automated treatment monitoring and image transfer was tested for feasibility of prospective use.

Results: For 129 patients, the SF automatically computed CBCTM for all fractions (2 min per fraction, 100% registered). Using MATLAB, CBCTM was only computed at fractions 10 and 20 (CBCTM10, CBCTM20) due to manual work requirements (55 min per fraction), with only 83/129 (64%) patients completing the registration process. For patients processed by both platforms, CBCTM10 and CBCTM20 were highly correlated (rho =0.81, 0.92, p < 0.005). Performance of logistic regression models combining CBCTM20 and mean lung dose to predict RP was similar (AUC of 0.678 for MATLAB, 0.678 for SF). For the 20 patients tested for automatic treatment monitoring of Mosaiq, 602 out of 613 fractions total had corresponding CBCTs in Mosaiq and 100% of the available images were successfully transferred to RayStation.

Conclusion: An SF was developed that enables real-time treatment monitoring with automated radiomic feature extraction. This SF successfully implemented a pipeline for deriving a CBCT marker for RP that is ready for clinical use in prospective studies.

Keywords

Computer Software, Feature Extraction, Image Processing

Taxonomy

IM/TH- Image Analysis (Single modality or Multi-modality): Computer-aided decision support systems (detection, diagnosis, risk prediction, staging, treatment response assessment/monitoring, prognosis prediction)

Contact Email

Additional Links