From the November 2022 issue of HealthCare Business News magazine
Jan Makela
Deep learning (DL) is literally transforming the field of imaging. Applications such as imaging interpretation and workflow simplification have already been widely discussed in the industry, but we are also finding enormous potential for DL algorithms in image reconstruction and acquisition. Here, we see a strong path to simultaneously improve image quality and reduce scan time compared to prior approaches.
DL-enabled advanced digital technologies are taking imaging to the next level of advanced capabilities by updating existing equipment with new software to provide clinicians with more robust, accurate, and data-driven information than ever before. By avoiding many of the shortcomings of traditional algorithms, DL is improving on previous versions of the iterative reconstruction (IR) imaging process.
A subset of machine learning, DL utilizes deep neural networks, which consist of layers of mathematical equations and millions of connections and parameters that get trained and strengthened based on the desired output. In doing so, it is making a significant leap forward in efficacy compared to previous processes that require more human intervention, because DL can handle complex models and vast numbers of parameters with ease.
Indeed, the industry has progressed beyond conventional IR and has entered an era of DL image reconstruction. And the breakthroughs in applying DL to imaging are quite stunning. For example:
• One emerging DL-based technology focusing on computed tomography (CT) imaging is aiming to solve the long-standing issue of having to choose between image quality and dose. Providing natural looking image quality, even at a low dose, this high-speed workflow can reconstruct a heart in less than a minute and an abdomen or pelvis in about 90 seconds.
• Similarly, with respect to magnetic resonance imaging (MRI), the DL-based image reconstruction algorithm helps address the age-old issue of having to compromise between image quality and scan time. The DL platform provides MR images that are reconstructed from raw data to achieve extraordinarily high-fidelity resolution, and can enable scan time reduction by up to 50%.
• And DL’s benefits are now expanding even further to positron emissions therapy-computed tomography (PET/CT) imaging, aiming to help provide clinicians image quality performance comparable to hardware-based Time-of-Flight, such as contrast-to-noise ratio and contrast recovery.
