Jonas Petersen

Thesis Topic: Development of a quality metric for MR image reconstruction using object detection

Supervisors: Vanya Saksena, Florian Knoll

Description

Magnetic Resonance Imaging is a widely used medical imaging modality due to its ability to depict structures and functions of tissues and organs without using ionizing radiation. However, due to its time-consuming nature, methods to increase data acquisition are used [1]. To ensure that the visible structures within the images are accurately depicted, a new evaluation metric is being created that incorporates object detection to identify and evaluate the medical content of the reconstructions.

To create this metric, the fastMRI, and fastMRI+ datasets are utilized [2,3]. For object detection, a Faster R – CNN model is used, which is a popular object detection algorithm that uses a region proposal network as well as a shared convolutional network to accurately detect objects in images [4]. To assess the quality of the reconstruction, the mAP is calculated by comparing predictions made using the reconstructed image against those using the original image.

[1] Anagha Deshmane, Vikas Gulani, Mark A Griswold, and Nicole Seiberlich. Parallel mr imaging. Journal of Magnetic Resonance Imaging, 36(1):55–72, 2012.
[2] Jure Zbontar, Florian Knoll, Anuroop Sriram, Tullie Murrell, Zhengnan Huang, Matthew J Muckley, Aaron Defazio, Ruben Stern, Patricia Johnson, Mary Bruno, et al. fastmri: An open dataset and benchmarks for accelerated mri. arXiv preprint arXiv:1811.08839, 2018.
[3] Ruiyang Zhao, Burhaneddin Yaman, Yuxin Zhang, Russell Stewart, Austin Dixon, Florian Knoll, Zhengnan Huang, Yvonne W Lui, Michael S Hansen, and Matthew P Lungren. fastmri+: Clinical pathology annotations for knee and brain fully sampled multi-coil mri data. arXiv preprint arXiv:2109.03812, 2021.
[4] Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real[1]time object detection with region proposal networks. Advances in neural information processing systems, 28, 2015.