Search

Improving on Workflow with Deep Learning Reconstruction

Dr. Mark Kon, Consultant Radiologist at the Bradford Royal Infirmary, UK, has worked with Canon Medical’s Aquilion Prime SP since the launch of the Second Edition and has seen the benefits from iterative reconstruction that are possible with Adaptive Iterative Dose Reduction (AIDR) 3D. Powered by Altivity, the Deep Learning Reconstruction (DLR) in Canon’s Advanced intelligent Clear-IQ Engine (AiCE) builds upon AIDR 3D technology to bring even greater gains in dose reduction, as well as image quality.

AiCE harnesses the enormous computational power of a Deep Convolutional Neural Network (DCNN) algorithm. It is trained to differentiate signal from noise, so that it can suppress noise and enhance signal. It can differentiate and produce images of high spatial resolution. Training involves advanced statistical Model Based Iterative Reconstruction (MBIR). However, unlike MBIR, AiCE Deep Learning Reconstruction is part of Canon Medical's Altivity suite of solutions, leveraging AI to overcome challenges of image appearance and/or reconstruction speed when used in a clinical setting.

“From my perspective, AIDR 3D already delivers low dose imaging, but AiCE builds upon this to provide better quality images for the same or even lower radiation dose,” remarked Dr. Kon.

An all-in-one integrated solution for everyday use

The process of DLR requires a lot of computing power and time - far more than would be available at the local CT workstation.

“For me, the intelligent part about AiCE is that once DLR has learned the reconstruction process, the much more compact based algorithm can be packaged, transferred and installed with the local CT scanner. DLR processing, therefore, becomes truly integrated with the CT system and does not require further access to remote supercomputers by direct connection, or through the cloud,” explained Dr. Kon. “This means that conventional CT rooms can accept Canon CT systems with AiCE as an all-in-one solution without depending on superfast cloud connectivity.”

“AiCE has effectively replaced filtered back projection as the reconstruction engine, and integrates into the background of all our scanning,” said Dr. Kon.

“We do not have to turn it on or off or select how much blending to choose. It is simply on all the time and built into every scanner and every examination, every day. It means that every examination takes advantage of DLR technology to achieve lower dose and better image quality.”

Image
sou

Training AiCE
The AiCE DLR is trained with high quality, advanced MBIR target images and learns to turn low-quality input data into low noise images that are sharp and clear. While in development, the AiCE DLR algorithm is taught to produce high signal-to-noise ratio (SNR) images through an intense training process. AiCE learns to differentiate signal from noise by training on selected, high-quality patient data sets that acquired with high tube current and reconstructed with all the benefits of state-of-the-art MBIR—including sophisticated system- and noise models, as well as a large number of iterations that are not possible clinically. As this time-consuming training process is completed before leaving the factory, the fully-trained AiCE DLR is able to work quickly in the clinic.

“With AiCE and a new PACS system, we only have to access one series of 1mm axials.”

Dr. Mark Kon, Consultant Radiologist at the Bradford Royal Infirmary, UK.

Noise-free images from thin slices

Dr. Kon has found the clear appearance of thin slice images one of the most remarkable benefits of AiCE in clinical practice. “Comparing AiCE 3 mm with AIDR 3D 3mm, there is much more noise in the AIDR 3D reconstruction,” he said. “It has not surprised us that AIDR 3D 1mm is noisier than 3mm slices. AiCE 1mm images demonstrate significantly less noise than AIDR 3D 1mm images, but what is truly impressive is that 1mm AiCE reconstructions are almost as noise-free as AiCE 3mm reconstructions. While this does not sound like a big step, it has a profound effect on how we store, reconstruct images, and view multi planar imaging.”

Image

Streamlining workflow and data

Previously, Dr. Kon and his team would use 1mm axial sections to reconstruct and save to PACS, but would have to review images at 3mm, due to noise. They also had to save separate series of 3mm axial, 3mm coronal, and 3mm sagittal reformats, as well as 1mm and 3mm lung images.

“With AiCE and a new PACS system, we only have to access one series of 1mm axials.We now view noise-free one millimeter axial images and have instant access to 1mm reformatted sagittal, and 1mm coronal images. And with AiCE, even the lungs can be viewed without resorting to a separate lung reconstruction,” he said. “This means a saving in PACS data storage, but more importantly, fewer series for the radiologist to open and close, improving on workflow.”

Image

Focused imaging

The low dose imaging achievable with AiCE enables Dr. Kon and his team to answer specific clinical questions with the lowest dose possible. “We use AiCE everyday for noise-free reconstruction, and to reduce radiation dose in every acquisition. Importantly, AiCE has enabled us to improve workflow for radiologists and has even encourages us to change the way we think by focusing our imaging on answering even more specific clinical questions,” said Dr. Kon. //
Image

Bradford Royal Infirmary, UK.