Classifying and Quantifying MRI Image Quality from DICOM Data at the Edge

Image quality is a critical aspect of diagnosis in radiology for the radiographic images produced by various scanner modalities. A radiologist can view and describe an image of good quality. Optimizing the image quality while minimizing the scan duration is crucial for operational efficiency and patient comfort. This paper presents a systematic approach to classifying MRI images and presenting them with quantitative attributes derived from the metadata of the DICOM (Digital Imaging and Communications in Medicine) images. We build MriqNet, a framework that executes deep learning algorithms on the DICOM images in a computing environment to determine the image quality and display the classified results in conjunction with DICOM header attributes. MriqNet quantifies the image quality, image scan duration, and other DICOM series-level metrics. We deploy and test MriqNet with our open-source software framework, Niffler, to run the image quality pipelines in environments like the cloud and the edge. Evaluations on MriqNet with MRI images show its accuracy and performance in classifying and quantifying image quality.