Implementation of Primary and Secondary Methods for Automatically Determining Necessity of Online Adaptive Replanning During MRI-Guided Adaptive Radiation Therapy
H Nasief*, A Parchur, E Omari, Y Zhang, X Chen, X Li, Medical College of Wisconsin, Milwaukee, WI
Presentations
SU-K-BRC-1 (Sunday, 7/10/2022) 5:00 PM - 6:00 PM [Eastern Time (GMT-4)]
Ballroom C
Purpose: Online adaptive replanning (OLAR) can be labor-intensive and time-consuming, and however, is not necessary for every fraction. We have recently showed that the necessity of OLAR for a given daily MRI set can be determined using machine leaning (ML) models based on either structural similarity index map (SSIM) or wavelet multiscale textures (WMT) extracted from thedaily MRI. This study aims to implement the two independent methods into a tool to quickly determine when OLAR is needed immediately after daily MRI acquisition during MRgART.
Methods: A graphical user interface (GUI) was developed using MATLAB® to implement the following procedure: (1) rigidly registering reference MRI to daily MRI immediately after its acquisition; (2) transferring 50 and 100% isodose surfaces created offline from the reference MRI onto the daily MRI; (3) calculating SSIM in the region of interest (ROI) enclosed by the 50 and 100% isodose surfaces; (4) extracting the engineered feature, MMF ratio=ln(Mean/(Moment×FWHM), from the SSIM map, and applying it to the SSIM classifier developed; (5) extracting three WMT features in the ROI and applying the three-feature combination to the WMT classifier developed; and (6) reporting the WMT-based prediction as the primary indicator and the SSIM-based as the secondary (validation) indicator on whether OLAR is necessary for the daily MRI. This process was fully automated and was tested using 20 daily MRI sets.
Results: The execution of the developed GUI was smooth and within 38 seconds using a hardware of Intel® Core™i7-6700 CPU @3.4GHz 32GB RAM x64-based processor. The independent primary (WMT) and the secondary (SSIM) predictions agreed for all the cases tested.
Conclusion: A newly developed GUI with the implementation of two independent machine learning classifiers can automatically and objectively determine if OLAR is necessity immediately after acquisition of a daily MRI and before recontouring, avoiding unnecessary effort during MRgART.
Keywords
Wavelets, Treatment Planning, MRI
Taxonomy
Not Applicable / None Entered.
Contact Email