The Molecular Magnetic Resonance Imaging and Machine Learning Laboratory at Tel Aviv University, funded by the Israel Science Foundation, is developing dedicated MRI sequences and algorithms that translate “signal signatures” into biological indicators – with a focus on multiple sclerosis and reducing dependence on contrast agents.
MRI is considered one of the most important imaging technologies in medicine, but in most cases it tells us what the tissue looks like – not what is happening in it at the molecular level. Dr. Or Perlman, a researcher at Tel Aviv University’s School of Biomedical Engineering and the Sagol School of Neuroscience, is trying to push this boundary. “We are developing new detection methods using MRI devices that provide insights into the human body and its diseases.” Dr. Perlman explains: “We want to transform MRI from an anatomical imaging tool into an early molecular diagnostic tool.
As it stands, most MRI scans provide basic structural and functional information: the shape of organs, the presence of tumors, edema or hemorrhage, and sometimes even mapping active areas in the brain. But at the cellular and molecular level, many diseases begin with subtle chemical changes in tissue composition long before a lesion appears that can be seen with the eye. This is where the idea of molecular imaging in MRI comes in. Using the same familiar instrument, with familiar physics, but with protocols designed in advance to be sensitive to the molecular environment and early metabolic processes.
The question is: Will molecular imaging in MRI contribute in the future to early diagnosis and a better understanding of the disease mechanisms of multiple sclerosis?
The process begins with the physics of the signal. Perlman and his team develop specific sequences of radio pulses and magnetic fields that excite the nuclear spins in the tissue in a way that depends heavily on their specific magnetic properties. Inflamed tissue, for example, will behave differently from healthy tissue; an area where cell density has changed or where metabolism is accelerated will produce a different signal signature. From the scanner’s perspective, the resulting signal is complex and noisy, and at first glance it looks like “just another MRI scan,” but inside it is a wealth of information about the state of the tissue. “We use the physics of the device to reveal not only structure but also processes.”
To decipher this complex signal, machine learning algorithms are incorporated into the research. The lab creates digital brain twins based on physical models, and also collects large volumes of data from preclinical disease models in which the “biological truth” is known – which areas are healthy, which are diseased, and to what degree. Machine learning systems learn to associate signal patterns with tissue status, building a kind of “dictionary” that translates complex MRI signatures into biological indicators. After such training, a new scan can be entered and a quantitative assessment of tissue characteristics can be obtained, which may indicate whether a disease process is beginning in a particular area, even when no clear structural changes are yet visible.
The combination of physics, machine learning, and preclinical experiments places MRI on the verge of a new role: not just a snapshot of “what it looks like,” but an early window into “what’s really happening” inside the tissue, even before a disease erupts in full force.
One of the practical goals of the project is to reduce the dependence on gadolinium-based contrast agents, which are currently common in some MRI examinations but raise safety questions, especially in repeated examinations. If it is possible to extract the molecular information from the signal of the water and tissues themselves, with the help of smart protocols and advanced algorithms, it will be possible to obtain an earlier and more accurate diagnosis without burdening the patient with additional contrast agents.
The research, funded by the National Science Foundation, focuses on multiple sclerosis and incorporates close collaboration with clinicians and brain researchers. The clinical questions are already defined at the planning stage: is the direction to identify small foci of inflammation, monitor response to treatment, differentiate between types of lesions, or early identification of degenerative diseases? At the same time, the limitations of using MRI are taken into account: reasonable scan time, patient comfort, and the need for clear output and not just complex maps for researchers. The goal is not only an impressive physical-computational achievement, but a process that can be integrated into clinical routine in the future.
In the long term, Perlman hopes that molecular imaging in MRI will become part of the toolbox of precision medicine: earlier diagnosis, tailoring treatment to the biological profile of the tissue, and a better understanding of disease mechanisms in the brain and other organs. The combination of physics, machine learning, and preclinical experiments places MRI on the verge of a new role: not just a snapshot of “what it looks like,” but an early window into “what’s really going on” inside the tissue, even before a disease erupts in full force.
The research is led by research student Ruth Ben-Haim and researcher Dr. Michal Rivlin.
More of the topic in Hayadan:
