Increasing the scanning time efficiency of magnetic resonance imaging systems through data integration and deep learning

点击翻译按钮获取中文摘要

Reconstruction of high-fidelity images from incomplete or noisy acquired data is a ubiquitous engineering challenge with many applications. Deep learning methods have become the state-of-the-art for image reconstruction. In this context, medical image reconstruction is essential for patient diagnostics and health management. However, these imaging exams are expensive, and wait times for imaging examinations are long and increasing in Canada. For example, for magnetic resonance imaging (MRI) scans, there is an average wait time of 9.3 weeks. These long MRI wait times are estimated to cost $700M/year to Canada's economy.Over 50% of new medical imaging exams are follow-up exams used for health screening and monitoring disease evolution and prognosis. Nevertheless, follow-up exams do not consider the information available from previous examinations during image acquisition and reconstruction. This research will focus on MRI exams, the diagnostic imaging modality with the longest wait times in Canada. We will develop a holistic solution to integrate the information of previous MRI exams to decrease the overall scanning time of follow-up exams. Although the patient's historical (past) imaging information is readily available through the existing healthcare infrastructure, the massive size of these imaging files hinders real-time (while the patients are still inside the MRI scanner) reconstruction of these images. Moreover, it is essential to ensure that the reconstructed images are not biased toward the previous MRI exam. Our goal is to reduce overall follow-up MRI scan times utilizing deep learning. Based on our feasibility studies, we expect to make follow-up exams up to four times faster. The successful implementation of this project will lead to millions of dollars in savings for Canada's healthcare systems by significantly reducing MRI wait times. Our methods can be easily extended to other imaging modalities, thus providing the landscape enabling more integrated and efficient imaging systems.