Deep Learning-Enhanced MRI for Assessing Myocardial Oxygenation

Advances in medical imaging technology have revolutionised cardiovascular diagnostics, yet non-invasive techniques for assessing myocardial oxygenation remain a challenge. Traditional methods like positron emission tomography (PET) are effective but have limitations such as radiation exposure, cost and accessibility. Recently, an innovative approach has emerged that incorporates deep learning with cardiovascular magnetic resonance imaging (CMR) to quantify myocardial oxygen extraction fraction (mOEF) and myocardial blood volume (MBV) without the use of contrast agents. This technique addresses existing limitations and holds promise for enhancing the diagnosis and monitoring of myocardial health, especially in conditions such as myocardial infarction.

Integration of Deep Learning with CMR

Introducing deep learning into the field of CMR is pivotal for improving image reconstruction and data accuracy. The method described involves a fully connected UNet-based neural network, termed DeepOxy, developed to generate mOEF and MBV maps. This system uses an asymmetric spin-echo prepared sequence integrated into a standard 3T MRI system. DeepOxy's unique capability lies in its training on synthetic data generated to simulate various physiological conditions, enabling the model to learn accurate estimations of myocardial oxygen extraction.

DeepOxy tackles persistent issues in traditional CMR imaging, such as signal distortion and inhomogeneity artefacts. By employing a theoretical model as its basis, the neural network decodes and reconstructs high-fidelity mOEF maps from MRI signals. This results in enhanced reproducibility and a significant reduction in artefacts, providing a reliable tool for assessing myocardial health.

Methodology and Reproducibility

The reproducibility of DeepOxy was rigorously assessed through a study involving 20 healthy participants aged 20 to 30. Each individual underwent CMR scans on two separate days, covering three short-axis slices of the left ventricle. The study evaluated the coefficient of variation (CV) to determine consistency. Findings showed a CV of 2.6% at the participant level, indicating high reproducibility. The mean global mOEF was 0.58, in line with established PET data for similar age groups, which typically report mOEF values of 0.60 to 0.68. The MBV measurements, with a mean of 9.5%, also aligned with values from comparable studies.

This reproducibility demonstrates that DeepOxy can be a dependable alternative to more invasive or complex methods like PET, providing a safer and quicker assessment. The study also revealed that while regional differences in mOEF were minimal among myocardial segments, female participants displayed slightly higher segmental mOEF than males, which warrants further investigation into gender-specific myocardial oxygen dynamics.

Clinical Implications and Future Applications

One of the most compelling aspects of DeepOxy-enhanced CMR is its potential application in clinical settings, particularly for patients with chronic myocardial infarction (CMI). In a pilot study involving ten patients, the method demonstrated its efficacy in identifying hypoxic regions within the myocardium. Results highlighted a marked reduction in mOEF in infarction cores compared to normal myocardial regions, underscoring the method’s sensitivity in detecting oxygenation deficits. Such capabilities are crucial for clinicians to evaluate the extent of myocardial damage and plan appropriate interventions.

Beyond infarction, this non-invasive tool could be transformative in managing other cardiac conditions like hypertrophic and dilated cardiomyopathy, heart failure, and even diseases with systemic impacts, such as chronic kidney disease. The quantitative assessment of mOEF and MBV offers a new dimension for tracking the progress of therapies to improve myocardial oxygenation and blood volume distribution. This could include evaluating novel treatments such as stem cell therapy, angiogenic factor delivery or other microcirculatory interventions.

The development of DeepOxy represents a significant advancement in non-invasive cardiac imaging. By leveraging deep learning, this CMR technique achieves reliable, reproducible and detailed quantification of myocardial oxygenation without the need for contrast agents or radiation exposure. The promising results from initial studies pave the way for broader clinical applications, enabling more effective diagnosis and monitoring of myocardial health. Future research should focus on expanding participant diversity, refining the technology for multi-slice imaging and validating findings through extensive clinical trials. Such advancements could solidify DeepOxy’s role as a cornerstone in non-invasive cardiology diagnostics.

Source: Radiology Advances

Image Credit: iStock