Neuroimaging as a Tool for Diagnosing and Tracking Multiple Sclerosis — BrainPost | Easy-to-read summaries of the latest neuroscience publications

The takeaway

Neuroimaging is a common and useful tool for diagnosing multiple sclerosis. Over the last 25 years, imaging technologies have improved to give us clearer pictures of the main features of MS at greater resolution. Further, machine learning algorithms are being used to advance our knowledge of MS.

MRI: a key tool for diagnosis

Multiple Sclerosis (MS) is a chronic and progressive autoimmune disorder that happens when the body’s immune cells invade the central nervous system and begin to attack myelin: a protective fatty tissue wrapped around the axons of neurons. Symptoms of MS can vary wildly, depending on the region under attack. If the optic nerve is attacked, people can experience loss of vision; if the spinal cord is attacked, shooting nerve pain can be the result.

Because MS symptoms are so variable, MRI is one of the key tools for diagnosis. MRI is non-invasive and allows a clinician to see lesions: areas of the brain that have been attacked by the body’s immune cells, appearing in the brain or spine even when the patient is not currently experiencing any symptoms. However, identifying the lesions in the first place can be difficult to the untrained eye, and being sure they are caused by MS (and not another disease) is also challenging. Therefore, to better diagnose and treat MS, we need to make it as easy as possible for clinicians to find and examine these lesions.

Using MRI to search for lesions

There are many types of MRI sequences, and each of them can highlight different characteristics of human tissue. MRI uses a strong magnetic field to align hydrogen nuclei (protons) so that they are oriented in the same direction. Then, it hits these protons with a radio frequency pulse that sends them all spinning. When the protons relax back into alignment, they emit radio energy that can be detected by the machine. Protons embedded in different types of tissue will have different resonant properties that affect how long they spin before relaxing. So, it is possible to optimize the radio pulse frequency and the time of detection in such a way that MS lesions will be more visible in the final image.

The current recommended MRI sequence for finding MS lesions is called T2 FLAIR. T2-weighted images use a slow radio pulse frequency and a longer wait time after a pulse for detection. These properties make the sequence excellent for picking up tissues with increased water content, whose hydrogen protons have a longer period of resonance. FLAIR stands for “Fluid Attenuated Inversion Recovery,” which uses an extra radio pulse to suppress signals coming from free-flowing fluids (water, cerebrospinal fluid), making those parts of the image less bright. MS lesions still show up brightly on the image.

Similar to FLAIR, other “inversion recovery” sequences have been developed (with names like PSIR and STIR), but these have not overtaken the FLAIR sequence for brain MRIs.

Most early imaging research in MS focused on the brain, but it is now recognized that getting a good spinal cord MRI can be important for a diagnosis. There can be lesions here as well, and they may cause more disabling symptoms. Spinal cord lesions can also help confirm a diagnosis of MS, ruling out other diseases known as MS “mimics” that look similar on brain MRIs. Spinal cord imaging is more prone to interference from bodily processes like heartbeat, breathing, and swallowing, so every advance in this field matters. In the spinal cord, the recommended MRI sequences are different, including STIR or PSIR mentioned above.

Another technique for lesion detection is to use a contrast agent called gadolinium, which is injected into the blood right before an MRI. It is called a contrast agent because it helps to enhance the contrast of the MRI signal wherever it goes, causing a bright glow on a scan. Normally, gadolinium cannot get through the blood-brain barrier (the tight network of cells that separates the central nervous system from the rest of the body). However, if a person is currently having an MS attack, the blood-brain barrier becomes leaky at the site of the lesion. So, the location of “active” lesions will glow brightly on the MRI. However, it is not ideal for the health of the patient to use gadolinium repeatedly since it can accumulate in the central nervous system.

How can artificial intelligence help?

MS is a field ripe for machine learning applications. The basic problem is one of image classification, which is a staple in the machine learning world. Some algorithms for lesion detection have recently been proposed, and these can identify lesions, calculate their volume, and measure overall brain size, These measures can also be tracked over time in patients to get a clearer idea of disease progression.

Many of these algorithms need to be fed a large set of training data—for MS, this means getting a huge number of correctly classified MRIs to learn from. If there are any systematic biases in our diagnosis of MS, these will be replicated by the machine learning program. Finally, someone ultimately must take responsibility for the diagnosis (a doctor, not a machine!), so adding artificial intelligence, while extremely useful, does complicate the accountability landscape.

Moving forward

MS is a disease of the central nervous system, and it has become easier to diagnose as our neuroimaging methods improve in this area of intense research. More advanced MRI sequences are on the horizon, even ones that can detect and quantify the myelin itself (the tissue under attack). Combined with powerful algorithms, these sequences offer clinicians much more information to draw on when making decisions about an MS diagnosis.

References +

Barkhof, F. (1997). Comparison of MRI criteria at first presentation to predict conversion to clinically definite multiple sclerosis. Brain, 120(11), 2059–2069.

Tintore, M., Rovira, A., Martınez, M. J., Rio, J., Dıaz-Villoslada, P., Brieva, L., Borras, C., Grive, E., Capellades, J., & Montalban, X. (2000). Isolated Demyelinating Syndromes: Comparison of Different MR Imaging Criteria to Predict Conversion to Clinically Deﬁnite Multiple Sclerosis. 5.

Bakshi, R., Ariyaratana, S., Benedict, R. H. B., & Jacobs, L. (2001). Fluid-Attenuated Inversion Recovery Magnetic Resonance Imaging Detects Cortical and Juxtacortical Multiple Sclerosis Lesions. Archives of Neurology, 58(5), 742.

Dolezal, O., Dwyer, M. G., Horakova, D., Havrdova, E., Minagar, A., Balachandran, S., Bergsland, N., Seidl, Z., Vaneckova, M., Fritz, D., Krasensky, J., & Zivadinov, R. (2007). Detection of Cortical Lesions is Dependent on Choice of Slice Thickness in Patients with Multiple Sclerosis. In International Review of Neurobiology (Vol. 79, pp. 475–489). Elsevier.

Thompson, A. J., Banwell, B. L., Barkhof, F., Carroll, W. M., Coetzee, T., Comi, G., Correale, J., Fazekas, F., Filippi, M., Freedman, M. S., Fujihara, K., Galetta, S. L., Hartung, H. P., Kappos, L., Lublin, F. D., Marrie, R. A., Miller, A. E., Miller, D. H., Montalban, X., Mowry, E. M., … Cohen, J. A. (2018). Diagnosis of multiple sclerosis: 2017 revisions of the McDonald criteria. The Lancet. Neurology, 17(2), 162–173.

Chen, Y., Haacke, E. M., & Bernitsas, E. (2020). Imaging of the Spinal Cord in Multiple Sclerosis: Past, Present, Future. Brain Sciences, 10(11), 857.

Wattjes, M. P., Ciccarelli, O., Reich, D. S., Banwell, B., de Stefano, N., Enzinger, C., Fazekas, F., Filippi, M., Frederiksen, J., Gasperini, C., Hacohen, Y., Kappos, L., Li, D., Mankad, K., Montalban, X., Newsome, S. D., Oh, J., Palace, J., Rocca, M. A., Sastre-Garriga, J., … North American Imaging in Multiple Sclerosis Cooperative MRI guidelines working group (2021). 2021 MAGNIMS-CMSC-NAIMS consensus recommendations on the use of MRI in patients with multiple sclerosis. The Lancet. Neurology, 20(8), 653–670.

Moazami, F., Lefevre-Utile, A., Papaloukas, C., & Soumelis, V. (2021). Machine Learning Approaches in Study of Multiple Sclerosis Disease Through Magnetic Resonance Images. Frontiers in Immunology, 12, 700582. https://doi.org/10.3389/fimmu.2021.700582

Martí-Juan, G., Frías, M., Garcia-Vidal, A., Vidal-Jordana, A., Alberich, M., Calderon, W., Piella, G., Camara, O., Montalban, X., Sastre-Garriga, J., Rovira, À., & Pareto, D. (2022). Detection of lesions in the optic nerve with magnetic resonance imaging using a 3D convolutional neural network. NeuroImage: Clinical, 36, 103187.