Artificial Intelligence Technology -Medical Imaging Has Been Greatly Improved
The quality of medical imaging has been greatly improved thanks to a new artificial intelligence technology.
Patient outcome is directly influenced by a radiologist's ability to make accurate diagnoses from high-quality diagnostic imaging investigations. Obtaining enough data to provide high-quality imaging, however, comes at a cost: greater radiation dose for computed tomography (CT) and positron emission tomography (PET) or excessively long scan periods for magnetic resonance imaging (MRI) (MRI). Now, researchers at Massachusetts General Hospital's (MGH) Athinoula A. Martinos Center for Biomedical Imaging have developed a new technique based on artificial intelligence and machine learning that allows clinicians to obtain higher-quality images without needing to collect additional data. In a report published today in the journal Nature, they describe the technique, which they call AUTOMAP (automatic transform by manifold approximation).
Artificial intelligence technology. | Medical Imaging
"Image reconstruction is an important aspect of the clinical imaging pipeline because it converts the raw data coming off the scanner into images that radiologists can interpret," explains Bo Zhu, PhD, a research fellow at MGH Martinos Center and first author of the Nature publication. "The traditional technique to picture reconstruction employs a chain of handcrafted signal processing modules that necessitate professional manual parameter tuning and are frequently unable to cope with flaws in the raw data, such as noise." We propose a novel paradigm in which deep learning artificial intelligence determines the best picture reconstruction algorithm automatically.
"We trained imaging systems to'see' the way humans learn to see after birth, not by directly programming the brain but by stimulating neural connections to adapt organically through repeated training on real-world instances," Zhu explains. "With this technique, our imaging systems can determine the optimum computational strategies to provide clear, correct images in a wide range of imaging settings automatically."
The method is a significant step forward in biomedical imaging. The researchers took advantage of recent advances in both artificial intelligence neural network models and the graphical processing units (GPUs) that drive the operations in developing it, because image reconstruction — particularly in the context of AUTOMAP — necessitates a massive amount of computation, especially during the training of the algorithms. Another critical component was the availability of massive datasets (sometimes known as "big data"), which are required to train large neural network models like AUTOMAP. The technique would not have been possible five years ago, or even one year ago, because it takes advantage of these and other improvements, according to Zhu.
Even beyond producing high-quality images in less time with MRI or with lower doses with X-ray, CT, and PET, AUTOMAP has a variety of potential clinical benefits. The approach could aid in making real-time judgments on imaging techniques while the patient is in the scanner due to its processing speed.
"Since AUTOMAP is built as a feedforward neural network, picture reconstruction takes only a few tens of milliseconds," explains senior author Matt Rosen, PhD, director of the MGH Martinos Center's Low-field MRI and Hyperpolarized Media Laboratory and co-director of the Center for Machine Learning. "To rebuild the images from some types of scanning, time-consuming computational processing is now required. When quick input is unavailable following first imaging, a repeat study may be necessary to properly identify a suspected anomaly. AUTOMAP would give real-time image reconstruction during scanning, allowing for better decision-making and maybe avoiding the need for future visits."
Notably, the technique could aid in the advancement of additional AI and machine learning applications. The current buzz about machine learning in clinical imaging is centered on computer-assisted diagnoses. AUTOMAP could help advance these technologies for future clinical applications because they rely on high-quality images for proper diagnostic evaluations.
"Our AI technique is shown amazing increases in accuracy and noise reduction, and so has the potential to advance a wide range of applications," Rosen explains. "We're ecstatic to be able to bring this into the clinical sector, where AUTOMAP can function in conjunction with low-cost GPU-accelerated computers to improve clinical imaging and results."
About the Author:
Joe Fanning is an app engineer who studied at Havard. He is CEO of JSearch, which is a high-quality search engine bot sort of like Google Search, but arguably better. Yes, that is the sub-headline.
Some of his clients include his friends. Greg, Enzo, and Zaki. Greg does Drum Tracking. Enzo is a Pro Wrestler. Check out his pro wrestling tees. Zaki is a sound recording engineer in North Bergen NJ. Coral Weber is a dancer and actress in NJ.
Patient outcome is directly influenced by a radiologist's ability to make accurate diagnoses from high-quality diagnostic imaging investigations. Obtaining enough data to provide high-quality imaging, however, comes at a cost: greater radiation dose for computed tomography (CT) and positron emission tomography (PET) or excessively long scan periods for magnetic resonance imaging (MRI) (MRI). Now, researchers at Massachusetts General Hospital's (MGH) Athinoula A. Martinos Center for Biomedical Imaging have developed a new technique based on artificial intelligence and machine learning that allows clinicians to obtain higher-quality images without needing to collect additional data. In a report published today in the journal Nature, they describe the technique, which they call AUTOMAP (automatic transform by manifold approximation).
Artificial intelligence technology. | Medical Imaging
"Image reconstruction is an important aspect of the clinical imaging pipeline because it converts the raw data coming off the scanner into images that radiologists can interpret," explains Bo Zhu, PhD, a research fellow at MGH Martinos Center and first author of the Nature publication. "The traditional technique to picture reconstruction employs a chain of handcrafted signal processing modules that necessitate professional manual parameter tuning and are frequently unable to cope with flaws in the raw data, such as noise." We propose a novel paradigm in which deep learning artificial intelligence determines the best picture reconstruction algorithm automatically.
"We trained imaging systems to'see' the way humans learn to see after birth, not by directly programming the brain but by stimulating neural connections to adapt organically through repeated training on real-world instances," Zhu explains. "With this technique, our imaging systems can determine the optimum computational strategies to provide clear, correct images in a wide range of imaging settings automatically."
The method is a significant step forward in biomedical imaging. The researchers took advantage of recent advances in both artificial intelligence neural network models and the graphical processing units (GPUs) that drive the operations in developing it, because image reconstruction — particularly in the context of AUTOMAP — necessitates a massive amount of computation, especially during the training of the algorithms. Another critical component was the availability of massive datasets (sometimes known as "big data"), which are required to train large neural network models like AUTOMAP. The technique would not have been possible five years ago, or even one year ago, because it takes advantage of these and other improvements, according to Zhu.
Even beyond producing high-quality images in less time with MRI or with lower doses with X-ray, CT, and PET, AUTOMAP has a variety of potential clinical benefits. The approach could aid in making real-time judgments on imaging techniques while the patient is in the scanner due to its processing speed.
"Since AUTOMAP is built as a feedforward neural network, picture reconstruction takes only a few tens of milliseconds," explains senior author Matt Rosen, PhD, director of the MGH Martinos Center's Low-field MRI and Hyperpolarized Media Laboratory and co-director of the Center for Machine Learning. "To rebuild the images from some types of scanning, time-consuming computational processing is now required. When quick input is unavailable following first imaging, a repeat study may be necessary to properly identify a suspected anomaly. AUTOMAP would give real-time image reconstruction during scanning, allowing for better decision-making and maybe avoiding the need for future visits."
Notably, the technique could aid in the advancement of additional AI and machine learning applications. The current buzz about machine learning in clinical imaging is centered on computer-assisted diagnoses. AUTOMAP could help advance these technologies for future clinical applications because they rely on high-quality images for proper diagnostic evaluations.
"Our AI technique is shown amazing increases in accuracy and noise reduction, and so has the potential to advance a wide range of applications," Rosen explains. "We're ecstatic to be able to bring this into the clinical sector, where AUTOMAP can function in conjunction with low-cost GPU-accelerated computers to improve clinical imaging and results."
About the Author:
Joe Fanning is an app engineer who studied at Havard. He is CEO of JSearch, which is a high-quality search engine bot sort of like Google Search, but arguably better. Yes, that is the sub-headline.
Some of his clients include his friends. Greg, Enzo, and Zaki. Greg does Drum Tracking. Enzo is a Pro Wrestler. Check out his pro wrestling tees. Zaki is a sound recording engineer in North Bergen NJ. Coral Weber is a dancer and actress in NJ.