Low-Field MRI, AI and Equity

Summary: Magnetic Resonance Imaging (MRI) is a sophisticated tool that provides three-dimensional detailed anatomical images without invasive procedures, crucial for detecting and diagnosing diseases. This technology leverages a strong magnetic field to align protons in the body, which are then disturbed by a radiofrequency current. As protons realign post-current, they emit energy detected by MRI sensors, enabling physicians to differentiate tissue types due to their magnetic properties. The precession of protons and their coherent phase during this process enhances the MRI signal, critical for detailed imaging.

MRI is invaluable in fields like neurology, oncology, and cardiology due to its superior soft tissue contrast and lack of ionizing radiation, presenting a safer alternative for repeated scans and vulnerable populations. Innovations like functional MRI and diffusion MRI further refine diagnostic capabilities, providing dynamic and detailed views of physiological and pathological processes.

The integration of Artificial Intelligence (AI) with MRI technologies, particularly in cancer detection, underscores a shift towards personalized medicine. AI enhances the accuracy of MRI by analyzing complex data sets, leading to tailored treatment strategies and improved diagnostic accuracy. However, the high cost and complexity of high-field MRI systems limit accessibility, prompting interest in lower-field MRIs. These systems, while less powerful, offer potential for wider use due to lower operational costs and simpler infrastructure requirements, making them particularly beneficial in resource-limited settings.

How does the precision of lower-field MRIs compare to high-field systems in clinical practice? What advancements might we see in the next decade that could bridge the gap between these technologies? Could AI-driven low-field MRIs improve healthcare delivery in underdeveloped regions?

What is an MRI and why is it important?

• Non-invasive imaging technology: Produces three-dimensional detailed anatomical images.
• Applications: Used for disease detection, diagnosis, and treatment monitoring.
• Magnetic properties: Uses strong magnetic fields to align protons, whose re-alignment and emitted energy upon returning to equilibrium are used to distinguish tissue types.

How does an MRI actually work?

• Magnetic alignment of protons: Protons align with the MRI’s magnetic field due to their magnetic moments.
• Net magnetization vector: A sum of all magnetic moments in the tissue, influenced by the magnetic field to create a predominant alignment.
• Precession and Larmor frequency: Protons precess around the magnetic field direction at a frequency proportional to the field's strength.
• RF pulse and energy absorption: An RF pulse at Larmor frequency causes protons to absorb energy and alter their spin states, disrupting their alignment and enhancing the MRI signal.
• Phase coherence: The RF pulse induces a temporary synchronization of proton spins, enhancing signal uniformity and strength.

Why do we actually use MRIs?

• High-resolution soft tissue imaging: Especially useful in neurological, musculoskeletal, cardiovascular, and oncological contexts.
• Safety: No ionizing radiation, making it preferable for repeated use and sensitive populations.
• Advanced techniques: Functional MRI and diffusion MRI offer detailed and dynamic imaging capabilities.

AI and MRI in breast cancer

• Advanced imaging techniques: Include Dynamic Contrast-Enhanced MRI and Automated Whole Breast Ultrasound.
• Role of AI: Enhances precision and personalization of diagnostics and treatments, integrating MRI data with genomic and clinical information.
• Radiogenomics: AI tools correlate MRI features with genomic data to predict outcomes and tailor treatments.

AI and MRI in prostate cancer

• Importance of segmentation: Critical for precise prostate volume estimation and planning of interventions.
• AI advancements: Improvements in machine learning models enhance segmentation accuracy, reducing variability and improving diagnostic outcomes.

Low field MRI to allow more people to get its benefits

• Accessibility and cost: Lower-field MRIs are less expensive and simpler, making them suitable for wider deployment, including in resource-limited settings.
• Technical considerations: Lower-field MRIs have reduced image quality but are becoming more viable with technological advancements.

What is an MRI and why is it important?

Magnetic Resonance Imaging (MRI) is a non-invasive imaging technology that produces three-dimensional detailed anatomical images. It is often used for disease detection, diagnosis, and treatment monitoring.?

MRI uses magnets which produce a strong magnetic field that forces protons in the body to align with that field. When a radiofrequency current is then pulsed through the patient, the protons are stimulated, and spin out of equilibrium, straining against the pull of the magnetic field. When the radiofrequency field is turned off, the MRI sensors are able to detect the energy released as the protons realign with the magnetic field.?

The time it takes for the protons to realign with the magnetic field and the amount of energy released changes depending on the environment and the chemical nature of the molecules. Physicians are able to tell the difference between various types of tissues based on these magnetic properties.

One of the main benefits of MRI is its ability to produce high-resolution images of soft tissues. This capability makes it especially useful in neurological (brain), musculoskeletal, cardiovascular, and oncological (cancer) imaging. Unlike X-rays or CT (computed tomography) scans, MRI does not use ionizing radiation, which can be harmful in high doses. As such, MRI is considered a safer alternative for imaging that requires multiple scans or when scanning vulnerable populations, such as pregnant women or young children.

In the case of the brain, it can differentiate between grey matter and white matter and can also be used to diagnose aneurysms and tumors. This is because MRI can capture high-contrast images that highlight abnormalities in the brain or body without requiring surgical intervention. MRI can also be used dynamically to track movements within the body, which is beneficial for both clinical and research purposes in fields like cardiology and orthopedics.

Other techniques, such as functional MRI (fMRI), which measures and maps the brain's activity by detecting changes in blood flow, or diffusion MRI, which? provides detailed images of the fibrous structures of the nervous system, help doctors have a better “mapping” of the tissues.

But, while MRI devices are very useful, they are also costly, with the magnet being the biggest expense, around 1 million $ per Tesla. Today's landscape is dominated by high-field MRI systems that, while potent and precise, are expensive and complex.??

And here is where lower-field devices come in.Software and hardware improvements are driving interest in lower-field devices. New scanners are being developed with simpler designs for easier production, maintenance, and use in places with limited resources. Some are portable, needing lightweight shielding. Also, there's less reliance on power-hungry gradient coils.?

Today's landscape is dominated by high-field MRI systems that, while potent and precise, are expensive and complex. Researchers are actively developing lower-cost, low-field MRI technology, striving for advancements that could make these machines more practical for everyday clinical use. Potential benefits include making MRI more accessible for brain and soft tissue imaging across various healthcare settings.?

How does an MRI actually work?

In the MRI's strong magnetic field, protons—nuclei of hydrogen atoms with a magnetic moment from their spin—align with the field direction. This creates a net magnetization vector aligned with the magnetic field, though alignment is not uniform across all protons.

Protons possess a property known as "magnetic moment" due to their inherent spin. The magnetic moment makes protons behave like tiny magnets themselves. In hydrogen atoms, which are abundant in the human body through water and fat, the nucleus is simply one proton, making these atoms particularly responsive in MRI settings.? When placed in a magnetic field, such as the one generated by an MRI machine, these "tiny magnets" (protons) tend to align with the direction of the magnetic field.?

The strong, uniform magnetic field created by the MRI aims to orient these proton spins in one direction. The extent to which the protons align along the magnetic field direction contributes to the overall magnetization of the tissue being examined.??

The "net magnetization vector" is a vector sum of all the magnetic moments of the protons in the sample. Ideally, in the absence of any external influence, half the spins would align in one direction and half in the opposite, resulting in no net magnetization. However, under the strong magnetic field of an MRI, more protons align with the field than against it, resulting in a net magnetization vector that is overall aligned with the magnetic field.

When placed in a magnetic field, these spinning protons begin to precess. Precession in this context refers to the motion of the axis of a spinning body (in this case, the proton) which describes a cone around the direction of the magnetic field (if this part seems confusing you can check this Youtibe video https://www.youtube.com/watch?v=uySdo9cFuVc)

The Larmor frequency is the specific frequency at which these protons precess in a given magnetic field. It is directly proportional to the strength of the magnetic field applied. Each type of nucleus in different magnetic fields will have its unique Larmor frequency.?

For hydrogen nuclei (protons), this frequency is crucial for MRI imaging. When the RF pulse is applied at the Larmor frequency, it introduces energy into the system that resonates with the spinning protons. Because the frequency of the RF pulse matches the natural frequency of the protons’ precession, the protons absorb this energy efficiently. This absorption causes a change in the spin state of the protons, making them move out of their initial alignment with the magnetic field.

This precise tuning is critical because it ensures maximum energy absorption by the protons, leading to greater disturbance in their alignment, which is necessary for producing a strong MRI signal. In essence, applying the RF pulse at the correct Larmor frequency maximizes the effectiveness of the MRI imaging process by optimally using the magnetic properties of the protons.?

This energy causes some of the protons to flip their magnetic alignment, essentially altering their spin states from one orientation to another (e.g., from aligning with the magnetic field to aligning against it).

This change disrupts the uniformity of the magnetic alignment, or net magnetization vector, which is the sum of all the individual magnetic moments of the protons in the field of view. As a result of the RF pulse, the net magnetization vector shifts away from its initial alignment along the magnetic field.?

One of the key effects of the RF pulse is to bring the protons into phase coherence. The RF pulse temporarily synchronizes their spins, causing them to precess in unison. This synchronization significantly enhances the uniformity and strength of the MRI signal that can be detected once the RF pulse is stopped. Phase coherence is crucial because the more coherent the phase of the protons, the stronger and clearer the resultant MRI signal will be.

The synchronized, coherent motion of the protons during and immediately after the RF pulse generates a stronger and more homogeneous signal, which is then detected by the MRI scanner’s sensors. This signal diminishes as the protons return to their lower energy state, realigning with the magnetic field and losing their phase coherence. The process of detecting the changes in the net magnetization vector as the protons return to equilibrium allows the MRI to create images that reflect the different types of tissues and their distinct properties.

AI and MRI in breast cancer

Advanced imaging techniques, such as Dynamic Contrast-Enhanced (DCE) MRI, Digital Breast Tomosynthesis (DBT), and Automated Whole Breast Ultrasound (AWBUS), have expanded the tools available to radiologists, allowing for more nuanced assessments tailored to individual patient profiles. Among these, breast MRI is particularly notable for its effectiveness in detecting breast cancer and evaluating tumor responses to treatment.

The integration of Artificial Intelligence (AI) with MRI is enhancing the precision and personalization of medical treatments. Breast cancer remains a significant health concern globally, with early detection being crucial for effective treatment and increased survival rates. Traditional methods like mammography, while useful, have limitations, particularly for those with dense breast tissue or those at high risk.

AI significantly enhances the analysis of complex datasets, combining MRI images with genomic, pathological, and clinical data to provide a comprehensive understanding of a patient's condition. This fusion of data supports personalized treatment approaches, shifting away from the generic "one-size-fits-all" methodology to more tailored strategies based on individual genetic makeup, lifestyle, and environmental factors.

AI applications in radiogenomics correlate MRI features with genomic assays to predict cancer recurrence. For example, multigene and microarray assays assess tumor expression profiles to estimate recurrence risks, while AI-driven tools like CADe (Computer-Aided Detection) and CADx (Computer-Aided Diagnosis) aid in quantifying image data and extracting valuable insights. Convolutional neural networks (CNNs) in AI can generate probability heatmaps that identify tumor regions linked to therapeutic responses, enhancing the predictive power of imaging studies.

AI and MRI in prostate cancer

Prostate MRI segmentation, a fundamental aspect of prostate imaging, holds pivotal clinical implications. It facilitates the precise estimation of prostate volume, a crucial parameter for calculating the serum prostate-specific antigen (PSA) density. Additionally, accurate segmentation is essential for MRI-guided biopsy procedures, such as transrectal ultrasound/MRI fusion guided biopsy systems, and for planning radiotherapy.

However, traditional manual segmentation methods are laborious and susceptible to interoperator variability, leading to inconsistent results. Recognizing this challenge, the integration of artificial intelligence (AI) has emerged as a promising solution. In recent years, the application of AI in prostate MRI segmentation has garnered significant attention, with several studies showcasing its potential to streamline and enhance this critical aspect of prostate cancer diagnosis and management.

Nowadays when we speak of AI we are usually speaking of machine learning, which is basically a method that helps computers learn from data to make predictions. Just like how we learn from experience. It helps computers make decisions and predictions without being told exactly what to do. It's like teaching a computer to get better at a task by practicing with examples. In healthcare, it's often used for precision medicine, where it predicts the most suitable treatments for patients based on their characteristics and treatment history. Unlike traditional ML methods, DL does not rely on pre-defined features, enabling it to capture complex relationships within data.

The mean Dice similarity coefficient is a metric commonly used to evaluate the performance of segmentation algorithms. It measures the overlap between the segmented region produced by the algorithm and the ground truth segmentation provided by manual annotations.

The mean Dice similarity coefficient is like a measuring stick we use to see how well a computer program, in this case, a deep learning model, is at drawing the outline of the prostate on MRI scans.

Imagine you have two drawings: one made by the computer program and the other drawn by a human expert. Now, you put these drawings on top of each other and see how much they overlap. The mean Dice similarity coefficient gives us a number between 0 and 1 that tells us how similar these two drawings are.

If the coefficient is closer to 1, it means the computer's drawing is very similar to the human's drawing, showing that the program did a good job. Which shows the potential of AI in streamlining these processes and increasing the speed while retaining the quality in the diagnosis.?

For instance, researchers have developed a DL model incorporating a three-dimensional (3D) fully convolutional network with deep supervision, achieving a mean Dice similarity coefficient of 0.88 compared to manual segmentations.?

Low field MRI to allow more people to get its benefits

As MRI technology progresses, the industry sees a shift towards more powerful machines that deliver sharper images quickly. However, the high costs and limited availability of these advanced MRI units contribute to disparities in healthcare access, with significant gaps in lower-income countries and even within affluent societies.?

Today's landscape is dominated by high-field MRI systems that, while potent and precise, are expensive and complex. Efforts to refine lower-cost, low-field MRIs are underway, aiming to make them more effective for clinical use. The emergence of more affordable, albeit less powerful low-field MRI machines, sparks a conversation about their integration into healthcare delivery, focusing on practical challenges and the quest for improved utility in clinical environments. Despite not matching the image quality of their high-strength counterparts, these machines present a viable option for broadening diagnostic reach.??

The interest in low field comes from the fact that they are less expensive than their high-field counterparts, not only in terms of initial purchase price but also in ongoing maintenance and operational costs.

High-field MRI devices are costly, with the magnet being the biggest expense, around 1 million $ per Tesla. And here is where lower-field devices come in.Software and hardware improvements are driving interest in lower-field devices. New scanners are being developed with simpler designs for easier production, maintenance, and use in places with limited resources. Some are portable, needing lightweight shielding. Also, there's less reliance on power-hungry gradient coils.?

Lower-field devices have been around but mostly for specific uses or have been discontinued. But recently, there's renewed interest, leading to FDA approval of several lower-field systems since 2018.They are often lighter, need less shielding, and some don't need cooling. They also have lower safety distances, so they can be closer to other machines or objects. This reduces setup costs and makes them more portable. Recently, vans with lower-field devices have been developed, making imaging more accessible. This approach has been successful in Japan.

What is exactly low field MRI?

MRI machines are classified based on their Tesla levels, with low-field MRI machines typically having a magnetic field strength below 1.5 Tesla. The range between 0.01T and 0.1T, as mentioned, represents very low magnetic field strengths used in certain specialized MRI machines. These low-field MRIs are distinguished from higher-field MRIs (1.5T and above, with 3T being common in many clinical settings) by their lower magnetic field strength, which impacts the machine's ability to produce detailed images.??

The difference between low-field and high-field MRI started in the 1980s. After 1985, when the first 1.5T scanners came, there was a big gap in citations. This gap grew during the 1990s and even more in the early 2000s. High-field scanners became popular because they offer better image quality, allowing faster imaging, less scan time,? higher resolution, better contrast, and advanced sequences.?

The stronger magnetic field aligns more protons, enhancing the net magnetization and thus improving the image quality. However, this also results in longer T1 relaxation times, more pronounced susceptibility to artifacts.?

T1 relaxation time refers to the time it takes for protons, which have been knocked out of alignment by an RF pulse, to realign with the magnetic field after the pulse is turned off. This relaxation process involves protons returning to their lower energy state, emitting energy as they do so, which the MRI system detects to form images. longer T1 relaxation times. This means it takes longer for protons to realign with the magnetic field, allowing for detailed imaging of certain tissue types, but also requiring more time per scan and increasing sensitivity to movement, which can affect image clarity.?

Higher energy levels are needed to excite protons at higher Larmor frequencies. While this results in images with higher signal-to-noise ratios (SNR) and better resolution, the higher energy can also lead to more pronounced susceptibility artifacts. These artifacts occur because different tissues and materials (like metal in medical implants or air in the lungs) have different magnetic susceptibilities, which can distort the magnetic field locally. This distortion leads to errors in the MRI image, particularly near the interfaces of these materials.

In contrast, low-field MRIs, which operate below 0.5 T, require RF pulses with less energy due to their lower Larmor frequency, resulting in lower power requirements and simpler, less costly RF equipment. The lower magnetic field strength means fewer protons are aligned, reducing the SNR and generally yielding images with lower resolution and less detail. Stronger magnets mean quicker scans. But recent tech advancements have improved image quality at lower-field strengths, making them useful in clinics. Advancements in signal processing and image reconstruction are continuously improving the quality of low-field MRI images.?

Low-field devices may not match high-field ones in image quality, but they could be especially useful in low-income countries, but it's important to consider technical support and expertise availability. Although with globalization and digitalization remote reading or automated systems could help with diagnoses where experts are scarce. While for now low field MRI scanners are not mainstream, the potential they hold to solve certain problems and the development of supporting software to deal with problems in image quality is starting to make them into an attractive alternative.

References:

Rinck, P. A. (2024). Magnetic Resonance in Medicine: A Critical Introduction (14th ed.). TRTF – The Round Table Foundation: TwinTree Media. Retrieved from www.magnetic-resonance.org.

Belue, M.J., & Turkbey, B. (2022). Tasks for artificial intelligence in prostate MRI. European Radiology Experimental, 6(1), Article 33. https://doi.org/10.1186/s41747-022-00287-9

Sheth, D., & Giger, M.L. (2020). Artificial intelligence in the interpretation of breast cancer on MRI. Journal of Magnetic Resonance Imaging, 51(5), 1310-1324. https://doi.org/10.1002/jmri.26878

Arnold, T.C., Freeman, C.W., Litt, B., & Stein, J.M. (2023). Low-field MRI: Clinical promise and challenges. Journal of Magnetic Resonance Imaging, 57(1), 25-44. https://doi.org/10.1002/jmri.28408