Time-Efficient Ultrasound Localization Microscopy Based on Novel Radial Basis Function Two-Dimensional Interpolation

Introduction

Ultrasound technology is a major medical imaging technique widely used for visualizing subcutaneous structures such as organs, muscles, and arteries due to its safety, cost-effectiveness, and non-invasiveness. However, the performance of traditional ultrasound imaging is limited by the diffraction limit, resulting in limited spatial resolution. When the frequency increases, the spatial resolution improves, but the penetration depth of the beam decreases, causing a trade-off between spatial resolution and penetration depth.

In the past decade, Ultrasound Localization Microscopy (ULM) has addressed the aforementioned trade-off problem. ULM generates super-resolved (SR) images by accurately localizing microbubbles (MBs) injected intravenously. These SR images provide essential information for understanding and diagnosing various diseases affecting vascular structure or blood flow, such as cancer, stroke, and arteriosclerosis. However, the current clinical application of ULM still faces two major obstacles: the need for low-density microbubbles and long data acquisition times, as well as high frame rate requirements.

To reduce data acquisition time and improve the feasibility of clinical applications, this study proposes a time-efficient ultrasound localization microscopy technique (teULM) based on novel radial basis function (RBF) two-dimensional interpolation. This technique aims to reduce the high frame rate requirement for obtaining super-resolved images through 2D interpolation. The effectiveness of this technique will be validated on multiple in vivo datasets.

Research Sources

The paper was completed in collaboration with the following authors: - Giulia Tuccio (Student Member, IEEE) - Sajjad Afrakhteh - Giovanni Iacca (Senior Member, IEEE) - Libertario Demi (Senior Member, IEEE)

The paper was published in the IEEE Transactions on Medical Imaging, Vol. 43, No. 5, in May 2024.

Research Workflow

a) Research Workflow

1. Data Acquisition: First, simulate low-frame-rate data acquisition by applying down-sampling (DS = 2, 4, 8, and 10) to high-frame-rate ULM data.

2. Data Interpolation: Perform up-sampling on the data using the proposed two-dimensional radial basis function (RBF) interpolation method to reconstruct missing frames.

3. Preprocessing and Filtering: Use Singular Value Decomposition (SVD) space-time filters in preprocessing to distinguish microbubble signals from surrounding tissue signals.

4. Microbubble Detection and Localization: Detect microbubbles based on high-intensity pixel values and localize them using a 2D multivariate Gaussian distribution method.

5. Microbubble Tracking and Accumulation: Track microbubbles frame-by-frame using a modified bi-partite tracking algorithm to ultimately obtain a super-resolved image.

b) Main Research Results

1. Datasets: Validate the effectiveness of teULM using four in vivo datasets (A: rat brain; B: rat kidney; C: rat tumor; D: rat brain embolism).

2. Results Evaluation: Compare SR images obtained through teULM with those from standard ULM. Results indicate that teULM can effectively restore vascular structures at lower frame rates (e.g., 100Hz). Quantitative evaluations of corresponding results were performed using various assessment metrics (Root Mean Square Error RMSE, Dice coefficient, retention trajectory percentage, and saturation).

3. Interpolation Method Effectiveness: Achieved the goal of restoring missing data, especially at higher down-sampling rates (e.g., DS=10).

c) Conclusion

The research shows that teULM can generate accurate SR images at a data acquisition frame rate an order of magnitude lower than standard ULM (from 1kHz to about 50Hz). This technique will aid the clinical application of ultrasound localization microscopy since standard clinical equipment has lower frame rates. Results confirm that when the sampling rate is reduced to 50Hz, teULM can still generate high-quality SR density maps; however, a slightly higher frame rate (250Hz) is needed to produce consistent velocity maps.

d) Research Highlights

• Innovative Method: The proposed teULM pipeline based on radial basis function two-dimensional interpolation significantly reduces the high frame rate requirement for obtaining super-resolved images.
• Extensive Validation: The new method’s effectiveness was validated through multiple in vivo datasets, showing its potential for application in complex physiological structures.
• Multiple Evaluation Metrics: The new method’s effects were comprehensively evaluated and validated using various metrics such as RMSE, Dice coefficient, retention trajectory percentage, and saturation.

e) Other Valuable Information

• Research Limitations: This study only tested two-dimensional datasets, and future expansions to three-dimensional datasets could further promote clinical applications.
• Movement Impact Not Discussed: For higher down-sampling rates, movement might cause errors in signal paths and target localization. Future research could consider incorporating motion compensation techniques.

Conclusion

This paper proposes a time-efficient ultrasound localization microscopy technique that achieves super-resolved imaging by reducing the data acquisition frame rate. Results indicate that high-quality SR density maps can still be produced when the frame rate is reduced to 50Hz. Although there is still a need for improvement in flow velocity estimation, this research provides a crucial basis for the clinical transition of ultrasound localization microscopy. Document and issue source marked.