Applying Artificial Intelligence to Mitigate Effects of Patient Motion or Other Complicating Factors on Image Quality

Abstract

Artificial intelligence, particularly deep learning, offers several possibilities to improve the quality or speed of image acquisition in magnetic resonance imaging (MRI). In this article, we briefly review basic machine learning concepts and discuss commonly used neural network architectures for image-to-image translation. Recent examples in the literature describing application of machine learning techniques to clinical MR image acquisition or postprocessing are discussed. Machine learning can contribute to better image quality by improving spatial resolution, reducing image noise, and removing undesired motion or other artifacts. As patients occasionally are unable to tolerate lengthy acquisition times or gadolinium agents, machine learning can potentially assist MRI workflow and patient comfort by facilitating faster acquisitions or reducing exogenous contrast dosage. Although artificial intelligence approaches often have limitations, such as problems with generalizability or explainability, there is potential for these techniques to improve diagnostic utility, throughput, and patient experience in clinical MRI practice.

AI MSK clinical applications: spine imaging

“…Hence, apart from traditional noise reduction strategies, e.g., scan time reduction in MRI, radiation dose reduction in CT, or more sophisticated statistical models, AI techniques have recently been investigated for noise reduction. To achieve this, mappings between noisy and noiseless images were trained, using natural images as well as images intentionally modified with added Gaussian noise of a defined range [27]. Subsequently, noise-free reconstructions from a noisy test image were successfully generated for MR, CT, and PET data [28][29][30].…”

“…The main components of a classical complete radiology service model are image acquisition, "read" images, report, and medical decision MRI is prone to motion artifacts due to comparably long acquisition times. While several methods of motion correction exist, DL-based solutions are sometimes preferred, as they can correct for motion during post-processing without the need of input information regarding degree or type of motion-induced image distortion [27].…”

“…Generally, tradeoffs must always be made between spatial resolution, signal-to-noise-ratio, and scan duration [27,31]. Reduced image quality is sometimes accepted in case of minor impact on diagnostic yield, since current routine post-processing includes different techniques for improving overall image quality, and therefore, the overall effect on image quality is minimal.…”

Super‐resolution head and neck MRA using deep machine learning

“…4 Nonetheless, further reduction of scan time would be desirable to improve patient comfort, reduce motion artifacts, hasten diagnostic evaluation, and compete with the shorter scan times of contrastenhanced CTA and MRA. [5][6][7] The 3D thin-slab QISS uses a stack-of-stars k-space trajectory with radial sampling performed in the transversal axis and phase-encoding performed in the slice direction. This kspace trajectory provides robustness to motion and pulsation artifacts as well as efficient imaging while maintaining spatial resolution and vessel sharpness.…”

“…Recently, 3D thin‐slab stack‐of‐stars quiescent interval slice‐selective (QISS) MRA has been shown to provide high spatial resolution of the entire neck and Circle of Willis in ≈7 minutes with better image quality than TOF 4 . Nonetheless, further reduction of scan time would be desirable to improve patient comfort, reduce motion artifacts, hasten diagnostic evaluation, and compete with the shorter scan times of contrast‐enhanced CTA and MRA 5‐7 …”