[ad_1]
A team of bio-engineers from the Polytechnic Institute of Rensselaer (RPI), with funding from the National Institute of Biomedical Imaging and Bioengineering (NIBIB), has developed a technique artificial intelligence (AI) using post-processing images to quickly convert computerized tomography to low dose. (CT) scans images of higher quality, compared to low-dose scans that do not use the AI technique. Computed tomography has become an imaging service commonly prescribed in modern medicine, providing a non-invasive, detailed and close-up view of internal anatomy and pathology. Low dose CT minimizes X-rays of the patient.
This hybrid, deep-learning, image-based image reconstruction technique integrates low-dose radiation tomography images with emerging neural network methods and delivers comparable images at a much higher speed than those produced with conventional methods. iterative reconstruction. Dr. Wang's team developed deep learning techniques for tomographic imaging and continued his research with the help of NIH to improve image quality and computer efficiency for low dose CT scanners. "
Behrouz Shabestari, Ph.D., NIBIB Program Director, Artificial Intelligence, Machine Learning and In-Depth Learning
With its increasing use, computed tomography contributes to 62% of the radiation dose imposed in the United States by all imaging modalities. Although the risk of developing cancer from such radiation exposure is low, the growing concern over the increasing use of CT makes the reduction of CT posology a clinical goal. Medical imaging engineers are working to develop technologies that reduce the radiation dose of the scanner without compromising its diagnostic performance.
CT scans are reconstructed from combinations of many X-rays taken from different angles. In their study published June 10, 2019, Nature Machine Intelligence, the team led by Ge Wang, Ph.D., Clark & Crossan Professor in the Department of Biomedical Engineering at RPI, and Manbadep Kalra, MD, Associate Professor of Radiology at Harvard Medical School and Radiologist at Mbadachusetts General Hospital have compared conventional methods of reconstructing images from commercial CT machines to a new method called Modularized Neural Network. . The new method is a type of artificial intelligence that researchers call machine learning or in-depth learning.
The modular neural network for computed tomography image reconstruction progressively reduces data noise so that radiologists can participate interactively in optimizing the reconstruction workflow. Each small increment of image quality improvement can be evaluated by radiologists based on the medical diagnosis they wish to make.
The researchers obtained low-dose tomodensitograms from 60 patients; 30 who described abdominal anatomy and the remaining 30 described thoracic anatomy. The scans represented three commercial CT products, all of which already used iterative image reconstruction algorithms (conventional approach) to reduce the noise of the image. Noise causes a reduction in image quality due to a low radiation dose scanner. The iterative reconstruction approach refers to the repeated steps that medical imagers are trying to generate to obtain computed tomography images consistent with some prior knowledge of imaging physics and image content. The researchers compared image reconstruction with the currently used iterative methods and their new deep neural network for image post-processing.
Three radiologists evaluated and recorded images for two functions: structural fidelity and noise suppression. Structural fidelity is the ability of the image to accurately describe anatomical structures in the field of view, which can be reduced by noise. The sound of the image appears as random patterns on the image that affect its clarity.
For the abdominal imaging, radiologists badigned higher scores to the images produced with the modular neural network method on two of the three readers and considered the images of the third device as of a quality comparable to those of the iterative reconstruction method. With regard to chest imaging, the experts found that the image quality was comparable between the two methods for all devices. Overall, the modular neural network yielded favorable or comparable results over the iterative method when radiologists evaluated structural fidelity and noise suppression.
The researchers add that their new method is much faster than current commercial methods and that institutions with current CT scanners of different brands can use their technique to produce similar image results. The results of the study confirm that deep learning could help produce high quality CT images at lower doses, while making this new approach much more effective than the iterative process, which is time consuming and time consuming. image noise artifacts.
The research was funded in part by a grant from NIBIB (EB017140) for research aimed at developing low-dose CT systems.
Source:
National Institute of Biomedical Imaging and Bioengineering (NIBIB)
Journal reference:
Shan, H. et al. (2018) Competitive performance of a modular deep neural network compared to commercial algorithms for reconstruction of low-dose CT images. Nature Machine Intelligence. doi.org/10.1038/s42256-019-0057-9.
[ad_2]
Source link
