WE‐G‐18C‐08: Real Time Tumor Imaging Using a Novel Dynamic Keyhole MRI Reconstruction Technique

Abstract

Purpose:To test the hypothesis that the novel Dynamic Keyhole MRI reconstruction technique can accelerate image acquisition whilst maintaining high image quality for lung cancer patients.Methods:18 MRI datasets from 5 lung cancer patients were acquired using a 3T MRI scanner. These datasets were retrospectively reconstructed using (A) The novel Dynamic Keyhole technique, (B) The conventional keyhole technique and (C) the conventional zero filling technique. The dynamic keyhole technique in MRI refers to techniques in which previously acquired k‐space data is used to supplement under sampled data obtained in real time. The novel Dynamic Keyhole technique utilizes a previously acquired a library of kspace datasets in conjunction with central k‐space datasets acquired in realtime. A simultaneously acquired respiratory signal is utilized to sort, match and combine the two k‐space streams with respect to respiratory displacement. Reconstruction performance was quantified by (1) comparing the keyhole size (which corresponds to imaging speed) required to achieve the same image quality, and (2) maintaining a constant keyhole size across the three reconstruction methods to compare the resulting image quality to the ground truth image.Results:(1) The dynamic keyhole method required a mean keyhole size which was 48% smaller than the conventional keyhole technique and 60% smaller than the zero filling technique to achieve the same image quality. This directly corresponds to faster imaging. (2) When a constant keyhole size was utilized, the Dynamic Keyhole technique resulted in the smallest difference of the tumor region compared to the ground truth.Conclusion:The dynamic keyhole is a simple and adaptable technique for clinical applications requiring real‐time imaging and tumor monitoring such as MRI guided radiotherapy. Based on the results from this study, the dynamic keyhole method could increase the imaging frequency by a factor of five compared with full k‐space methods.