Efficient Microbubble Trajectory Tracking in Ultrasound Localization Microscopy Using a Gated Recurrent Unit-Based...

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Artificial Intelligence

Background:

• Ultrasound localization microscopy (ULM) offers noninvasive microvascular imaging but faces challenges with traditional microbubble tracking.
• Existing methods are complex, hindering real-time applications, and deep learning approaches often ignore temporal dynamics.

Purpose of the Study:

• To develop a novel deep learning model for efficient and accurate microbubble trajectory tracking in ULM.
• To address limitations in real-time processing and temporal motion modeling of microbubbles.

Main Methods:

• Introduced a gated recurrent unit-based multitasking temporal neural network (GRU-MT) for simultaneous trajectory tracking and optimization.
• Enhanced a nonlinear motion model to better capture microbubble dynamics.
• Evaluated various temporal neural networks (RNN, LSTM, GRU, Transformer) for microbubble tracking.

Main Results:

• GRU-MT demonstrated superior nonlinear modeling and robustness in both simulated and in vivo datasets.
• The proposed method achieved reduced trajectory tracking errors, especially in shorter time intervals.
• GRU-MT showed potential for efficient real-time microbubble trajectory tracking.

Conclusions:

• GRU-MT offers a significant advancement in microbubble trajectory tracking for ULM.
• The model's ability to handle nonlinear motion and temporal dynamics enhances tracking accuracy and efficiency.
• Open-sourced code facilitates further research and application of this technique.