Image: MRI images reconstructed from the same data with conventional approaches (L) and AUTOMAP (R) (Photo courtesy of MGH).
A new technique based on artificial intelligence (AI) and machine learning could enable clinicians to acquire high-quality images from limited data.
Developed at Massachusetts General Hospital (MGH; Boston, USA), the new image manipulation technique, called automated transform by manifold approximation (AUTOMAP), offers a unified framework for image reconstruction by recasting it as a data-driven supervised learning task, which allows mapping between the sensor and the image domain to emerge from an appropriate body of training data. To develop AUTOMAP, the researchers took advantage of the many advances made in neural network models used for AI.
Improvement in graphical processing units (GPUs) that drive the operations also contributed to the powering of image reconstruction algorithms such as AUTOMAP, as they require an immense amount of computation, especially during the training phase. Another factor was the availability of large datasets--known as big data--that are needed to train large neural network models. The overall result is a superior immunity to noise and a reduction in reconstruction artefacts, compared with conventional handcrafted reconstruction methods.
AUTOMAP also offers a number of potential benefits for clinical care, even beyond producing high-quality images in less time with magnetic resonance imaging (MRI) or with lower doses with X-ray, computerized tomography (CT) and positron emission tomography (PET). Because of its processing speed, the technique could help in making real-time decisions about imaging protocols while the patient is still inside the scanner. According to the researchers, the AUTOMAP technique would not have been possible five years ago, or maybe even one year ago. The study was published on March 21, 2018, in Nature.
“The conventional approach to image reconstruction uses a chain of handcrafted signal processing modules that require expert manual parameter tuning’ and often are unable to handle imperfections of the raw data, such as noise,” said lead author Bo Zhu, PhD, of the MGH Martinos Center for Biomedical Imaging. “With AUTOMAP, we've taught imaging systems to 'see' the way humans learn to see after birth, not through directly programming the brain but by promoting neural connections to adapt organically through repeated training on real-world examples.”
“Since AUTOMAP is implemented as a feedforward neural network, the speed of image reconstruction is almost instantaneous, just tens of milliseconds,” said senior author Matt Rosen, PhD, of the center for machine learning at the MGH Martinos. “Our AI approach is showing remarkable improvements in accuracy and noise reduction and thus can advance a wide range of applications. We're incredibly excited to have the opportunity to roll this out into the clinical space where AUTOMAP can work together with inexpensive GPU-accelerated computers to improve clinical imaging and outcomes.”
Deep learning is part of a broader family of AI methods based on learning data representations, as opposed to task specific algorithms. It involves artificial neural network (ANN) algorithms that use a cascade of many layers of nonlinear processing units for feature extraction and transformation, with each successive layer using the output from the previous layer as input to form a hierarchical representation.
Massachusetts General Hospital