Over almost two decades working in private clinical radiology with a focus on musculoskeletal imaging, I’ve seen an astounding progressive evolution in imaging technology. Until recently, advances in image quality have largely been a reflection of the steady evolution of engineering technology such as the emergence of high magnetic field strength MRI and its widespread adoption in clinical practice.

But in the 2020’s, game-changing leaps forward in MRI image resolution are happening, riding on the wave of the rapidly evolving space of Artificial Intelligence and deep learning reconstruction image enhancement technology. A long-held concept in radiology is that you never get something for nothing – ie – improvements in image quality must come at the cost of upgrades in engineering technology. But with AI, the rule book appears to have gone out the window – deep learning reconstruction paves the way for enhanced image quality with less noise, with no engineering upgrades… did we just get something for nothing? Almost.

We stand at the very beginning of a long and promising journey as we move into the age of AI in medical imaging. It’s early days in many ways though. I personally believe machine learning has a long way to go in the space of image analysis before it can add robust, tangible value to patient outcomes. But advancements in the AI image enhancement space are far more promising, with the emergence of deep learning algorithms that add significant and measurable improvements to image quality and noise reduction in both CT and MRI imaging. (1) .

Deep Learning Reconstruction has shown to quantitatively improve image sharpness and fine detail, reduce noise (improved SNR) and image artefacts, particularly fine motion blur artefact. In musculoskeletal imaging, with DLR technology we have seen a quantum step forward in image quality for high resolution 3T MRI, particularly for fine structures such as tiny ligaments and tendons of the wrist, hand and foot.

3T MRI with DLR is adding diagnostic value over conventional 3T MRI in a number of areas, particularly in relation to the diagnosis of injury to small parts such as the hands and feet. Some common examples encountered in routine clinical practice include tears of the scapho-lunate ligament (a commonly missed diagnosis, particularly in FOOSH type injuries), injury to the small pulleys and tendons of the fingers, and distinguishing tears of the ulnar collateral ligament (UCL) of the thumb (skiier’s thumb) from Stener lesions (UCL tear with displacement of the torn ligament superficial to the adductor aponeurosis, which may delay healing).

In imaging of the feet 3T MRI with DLR is shining a brighter light on fine structures that are often not clearly defined on conventional 3T MRI, such as injury to the bands of the spring ligament, tears of the LisFranc ligament complex and other fine ligaments of the ankle and foot, defining plantar plate tears, turf toe type injury and small Morton’s neuromas.

These are all important clinical diagnoses to make from a management and prognosis perspective, and some of these conditions may have suboptimal outcomes if not detected early.

From a quantitative perspective, whilst it is still early days for Deep Learning Reconstruction image enhancement technology, the early data is showing there is a clear and measurable improvement in image quality, reduced noise artefact and faster scan times, which is good news for the significant proportion of the community who experience claustrophobia or perhaps just general uneasiness in the MRI tunnel.

The table below demonstrates the difference in image quality between standard 3T MRI vs 3T MRI with DLR fast scanning at double to triple speeds, with a notable significant improvement in overall image quality by around 25%. (3)

The above is a sample image comparison between conventional 3T MRI, double-speed 3T MRI with DLR, and triple-speed 3T MRI with DLR. (3)

As a musculoskeletal radiologist, it’s an exciting time to be alive as we ride the metaphorical wave of advances in the AI image enhancement space.

And as a shameless tech geek, I’m passionate about exploring advances in imaging technology that offer the potential to improve clinical outcomes.  For this reason at PRS we’ve invested in a wide range of leading edge imaging technology, including 3T MRI and low dose CT with AI image enhancement.

1. Dana J. Lin MD, Patricia M. Johnson PhD, Florian Knoll PhD, Yvonne W. Lui MD; Artificial Intelligence for MR Image Reconstruction: An Overview for Clinicians; Journal of Magnetic Resonance Imaging; Feb 2020

2.   Judith Herrmann  , Gregor Koerzdoerfer, Dominik Nickel , Mahmoud Mostapha,  Mariappan Nadar, Sebastian Gassenmaier, Thomas Kuestner, Ahmed E Othman ; Feasibility and Implementation of a Deep Learning MR Reconstruction for TSE Sequences in Musculoskeletal Imaging; Diagnostics; Aug 2021

3.  Nobuo Kashiwagi, Hisashi Tanaka, Yuichi Yamashita, Hiroto Takahashi, Yoshimori Kassai, Masahiro Fujiwara, Noriyuki Tomiyama; Applicability of deep learning-based reconstruction trained by brain and knee 3T MRI to lumbar 1.5T MRI; Acta Radiologica; 2021