MRI-based automated detection of implanted low dose rate (LDR) brachytherapy seeds using quantitative susceptibility mapping (QSM) and unsupervised machine learning (ML). Issue 3 (December 2018)

• Record Type: Journal Article
• Authors: Nosrati, Reyhaneh; Soliman, Abraam; Safigholi, Habib; Hashemi, Masoud; Wronski, Matthew; Morton, Gerard; Pejović-Milić, Ana; Stanisz, Greg; Song, William Y.
• Is Part Of: Radiotherapy and oncology. Volume 129:Issue 3(2018)
• Page Start: 540
• Page End: 547
• Publication Date: 2018-12
• Subjects: MRI-only seeds detection -- Quantitative susceptibility mapping -- Machine learning -- Post implant dosimetry -- Prostate permanent seed brachytherapy; Oncology -- Periodicals; Radiotherapy -- Periodicals; Tumors -- Periodicals; Medical Oncology -- Periodicals; Neoplasms -- radiotherapy -- Periodicals; Radiotherapy -- Periodicals; Radiothérapie -- Périodiques; Cancérologie -- Périodiques; Tumeurs -- Périodiques; Electronic journals; 616.9940642
• DOI: 10.1016/j.radonc.2018.09.003
• Languages: English
• ISSNs: 0167-8140

Abstract

Highlights: Quantitative Susceptibility Mapping generated positive contrast for brachytherapy seeds. Seeds were differentiable from calcifications on magnetic susceptibility maps. An unsupervised machine learning algorithm was developed for seed detection and localization. The output of the proposed MRI-based workflow was in perfect agreement with the CT-based approach. Abstract: Background and purpose: Permanent seed brachytherapy is an established treatment option for localized prostate cancer. Currently, post-implant dosimetry is performed on CT images despite challenging target delineation due to limited soft tissue contrast. This work aims to develop an MRI-only workflow for post-implant dosimetry of prostate brachytherapy seeds. Material and methods: A prostate mimicking phantom containing twenty stranded I-125 dummy seeds and calcifications was constructed. A three-dimensional gradient-echo MR sequence was employed on 3T and 1.5T MR scanners. An optimized quantitative susceptibility mapping (QSM) technique was applied to generate positive contrast for the seeds and calcifications. Seed numbers, centroids, and orientations were determined using unsupervised machine learning algorithms (K-means and K-medoids clustering). The geometrical seed positions and the resulting dose distribution were compared to the clinical CT-based approach. Results: The optimized QSM-based method generated high quality positive contrast for the seeds that were significantly different from that for calcifications and could be easily differentiated by thresholding. The estimated seed centroids from both 3T and 1.5T MR data were in perfect agreement with the standard CT-based seed detection algorithm (maximum difference of 0.7 mm). The estimated seed orientations were highly correlated with the actual orientations ( R > 0.98). Conclusions: The proposed MRI-based workflow enabling an accurate and robust means to localize the seeds (position and orientation) upon validation on complex seed configurations, has the potential to replace the current widely practiced CT-based workflow.