A Better Map
By Beth W. Orenstein
Radiology Today
Vol. 26 No. 6 P. 22
Ultrafast, high-resolution MRI scans can potentially spot disease before symptoms occur.
Understanding how the brain works and what happens when it is injured or diseased is one of the most challenging endeavors of our time, says Zhi-Pei Liang, PhD, a professor of electrical and computer engineering and a member of the Beckman Institute for Advanced Science and Technology at the University of Illinois Urbana-Champaign (UIUC).
That is why Liang and colleagues are excited about their work, reported in the June 20 journal Nature Biomedical Engineering, which uses clinical MRI machines to image brain metabolism in a way that could give researchers and clinicians unique insight into brain function and disease.
For the last 40 years, MRI has played a major role in unlocking the mysteries of the brain. Conventional MRI provides high-resolution, detailed imaging of structures in the brain. Functional MRI (fMRI) is used to detect changes in blood flow and blood oxygenation levels, which are closely linked to neurological activity. However, neither provides information on the metabolic activity in the brain, which is important for understanding function and disease. Metabolic and physiological changes in the brain, as elsewhere in the body, often occur before structural and functional abnormalities can be seen on conventional MRI and fMRI, Liang says.
MRSI
Magnetic resonance spectroscopic imaging (MRSI) offers better detection and diagnosis of brain diseases; however, earlier MRSI methods took too long to capture images from brain metabolites and neurotransmitters in high spatial resolution and signal-to-noise ratio. The UIUC researchers were able to significantly improve MRSI data acquisition and processing to reduce the scan time for whole-brain imaging to 12.5 minutes.
“Our technology overcomes several long-standing technical barriers to fast, high-resolution metabolic imaging by synergistically integrating ultrafast data acquisition with physics-based machine learning methods for data processing,” Liang explains.
Conventional MRI and fMRI capture MR signals from water molecules. In addition to water molecules, MRSI measures signals from brain metabolites and neurotransmitters— N-acetyl aspartate, creatine, choline, myo-inositol, lactate, glutamine (GLN), glutamate (GLU), and gamma-aminobutyric acid (GABA). “The challenge lies in how to separate MR signals from different molecules at each spatial location,” Liang says.
MRSI is achieved by using special data acquisition and processing schemes, Liang says. For data acquisition, MRSI encodes both spatial and spectral information of multiple molecules into the MR signals generated using radiofrequency excitations. For data processing, the measured MR signals are decoded to determine the molecular contents at each spatial location.
MRSI is also known as spatially resolved MR spectroscopy (MRS). “So, MRSI can be viewed as an extension of MRI with spectroscopy capability or an extension of MRS with imaging capability,” Liang says. “Personally, I think MRSI is an ingenious integration of MRI with MRS. It provides both anatomical and biochemical information about tissues, revealing details about the chemical composition and metabolic activity within the tissues.”
Reducing Scan Time
Liang says the integrated approach allowed his team to overcome several technical hurdles. Here is what they did to reduce the scan time:
Hybrid Free Induction Decay and Spin Echo Acquisitions
By combining these two MRSI signal acquisition methods, the researchers were able to enhance data acquisition efficiency and improve the accuracy of separating different molecules.
Rapid Scanning With Short TR and Sparse Sampling (k,t)-Space
This involved using short repetition times (TR), which directly impacts the speed of data acquisition, and sparsely sampling k-space (the raw data acquired during an MRI scan), but doing so in a way that minimizes information loss. Advanced processing techniques then reconstruct images from sparsely sampled data.
Elimination of Water and Lipid Suppression Pulses
Traditional MRSI often requires suppressing the strong signals from water and fat to better visualize the signals of interest. The new technique eliminates this step, making it simpler and faster for clinical use. Incorporation of Navigators for Subject Motion Detection and Tracking B0 Field Drift Subject motion can severely degrade image quality in MRI, and variations in the magnetic field (B0 field drift) can also introduce artifacts. By embedding “navigators” into the sequence, the researchers were able to detect and correct for these factors, improving image quality without extending scan time.
Subspace-Based Machine Learning With Physics-Based and Data-Driven Priors
This advanced data processing technique leverages machine learning to reconstruct high-quality images from the sparsely sampled and noisy data. It utilizes both established physics principles and empirical data patterns to guide the reconstruction process, enabling robust image generation even with limited acquired data.
Identifying Potential
It is well-known that N-acetyl aspartate, myo-inositol, choline, creatine, GLU, GLN, lactate, GABA, glutathione, taurine, and N-acetylaspartylglutamate provide important biomarkers for neuronal integrity, glial proliferation, cell membrane turnover, astrocytosis (astrocytes are a type of glial cell in the central nervous system), inflammation, hypoxia, and excitatory/inhibitory synaptic neurotransmission.
“So, these molecular markers can help improve the sensitivity and specificity in detecting and characterizing a number of neurological disorders,” Liang says.
The researchers tested their MRSI technique on several populations, including healthy subjects, patients with brain tumors, and patients with multiple sclerosis. In healthy subjects, they were able to find and map varying metabolic and neurotransmitter activity across different brain regions, indicating that such activity is not universal.
In people with brain tumors, they found metabolic alterations such as elevated choline and lactate in tumors of different grades. They were able to differentiate tumor types based on the metabolic images, even though these tumors appeared identical on clinical MRI images. In subjects with multiple sclerosis, the technique detected molecular changes associated with neuroinflammatory response and reduced neuronal activity up to 70 days before changes became visible on clinical MR images, the researchers report. The researchers also obtained encouraging results from stroke, epilepsy, and Alzheimer’s disease in their ongoing studies.
“We can imagine that fast, highresolution metabolic imaging can be integrated into existing clinical scan protocols to improve diagnosis and treatment planning of neurological disorders,” Liang says. “If further research can confirm that the molecular markers obtained using our technology, in combination with blood-based markers, cerebrospinal fluid-based markers, and genetic markers, it could lead to early detection of neurodegenerative diseases (such as Alzheimer’s disease). It could also change the current radiological practice of imaging neurodegenerative diseases.”
Inspiration
Liang says the idea for this research came from his UIUC postdoctoral advisor, Paul Lauterbur, PhD, a chemist who, while at Stony Brook University, published the first true MR image in Nature in March 1973. Lauterbur had the idea for MR spectroscopic imaging soon after he published his 1973 paper on MRI, which won him the Nobel Prize. Lauterbur published perhaps the first MR spectroscopic imaging paper in 1975 in the Journal of the American Chemical Society, although that paper is not well-known, especially to young researchers, Liang says.
“Paul also envisioned the possibility and potential of fast, high-resolution MR spectroscopic imaging,” Liang says. “He asked me to work on the problem in the early 1990s when I was his postdoc. So, it has been a long journey for me and my students.” He adds that, over the years, many of his MR colleagues have tackled the problem and done outstanding work to advance the field of MR spectroscopic imaging.
In addition, MR manufacturers have significantly improved the performance of MRI scanners, which laid a solid foundation for Liang’s team to achieve the fast, high-resolution brain metabolic imaging using MR spectroscopic imaging reported in their paper. “Currently, all clinical MRI scanners are capable of performing fast, high-resolution MR spectroscopic imaging experiments, as long as they have the right data acquisition sequences and processing tools installed,” Liang says.
Exciting Work
Fei Du, PhD, director of the Laboratory for High-Field Imaging and Translational Neuroscience at McLean Hospital at Harvard Medical School, is equally excited and impressed with the Liang team’s work. Du calls the work “remarkable” and “groundbreaking.” He says it “has broad implications for metabolic neuroimaging and precision medicine, particularly for psychiatric disorders. It stands out for several innovations, pushing the boundaries of clinical MRSI and aligning with the growing paradigm of precision medicine.”
Achieving subcentimeter resolution in whole-brain metabolic mapping at 3T has historically been challenging, due to limited sensitivity and spectral resolution. Du says, “Liang’s team demonstrates that high-quality neurometabolic imaging is now within reach of standard clinical and research environments, which is crucial for translational neuroscience research and clinical applications.”
Spectral overlap at 3 T makes it difficult to reliably quantify neurotransmitters such as GABA, GLU, and GLN. “This study employs an optimized data acquisition strategy, J-resolved spectral editing, and advanced postprocessing to disentangle GABA from other J-coupled resonances,” Du says. “This is crucial for imaging excitatory-inhibitory balance in neuropsychiatric disorders, particularly for epilepsy and schizophrenia—an achievement long considered prohibitively difficult without ultrahigh-field systems.”
By combining physics-informed machine learning with mathematical modeling to accelerate data acquisition and imaging reconstruction, Du adds, this multidisciplinary approach substantially reduces scan times and improves quantification accuracy, taking an essential step toward clinical implementation. He has one caution, however. “Although the study includes validation across diverse clinical cohorts and sites, larger-scale studies in populations with psychiatric or neurodegenerative disorders are needed to confirm the specificity and clinical relevance of the reported metabolic signatures,” Du says. For example, he notes that full separation of GABA from glutathione, GLU, and GLN remains a technical challenge at 3 T.
Importance of Play
Liang says MR is a fundamental physical phenomenon that provides a unique window into biology. “It has never ceased to amaze us, in terms of what it can do. I think Erwin Hahn, the discoverer of the spin echo phenomenon, said it the best, ‘There is nothing that nuclear spins will not do for you, as long as you treat them as human beings.’” Liang says his team will continue to “play” with MR to develop advanced imaging technology that improves health care, especially precision medicine.
“In the short term,” he says, “we will leverage our advances in ultrafast, high-resolution metabolic imaging without water suppression and the significant progress in machine learning to develop an AI-powered MRSI technology that will allow us to map brain structures, function, and metabolism simultaneously. We are already getting very encouraging results in this direction and hope such a capability would significantly enhance our ability to detect and characterize neurological and neurodegenerative diseases and assess their therapeutic effects.”
Liang believes that being able to track metabolic changes over time will open many opportunities for research and clinical applications, not only in the brain but also elsewhere in the body. “As an engineer,” he says, “I don’t have sufficient science and clinical expertise to make good predictions on potential science and clinical applications of metabolic imaging.”
However, he would “venture to say that longitudinally tracking metabolic changes may help clinicians better monitor the efficacy of therapeutic treatments and allow timely adjustments of treatments to achieve personalized, precision medicine. I also can imagine that metabolic imaging can help build better predictive physiology models, also known as digital twins, for disease progression.”
—Beth W. Orenstein of Northampton, Pennsylvania, is a regular contributor to Radiology Today.