RA-BiMENet: Continuous-Time 4D Medical Image Interpolation via Relation-Aware Bi-Directional Motion Estimation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Four-dimensional (4D) medical imaging enhances 3D spatial data with a temporal dimension for dynamic organ motion analysis.
• Challenges in 4D imaging include low temporal resolution, insufficient sampling, motion artifacts, and blurring, hindering accurate diagnosis.
• High-quality, temporally continuous intermediate frame generation is crucial for patient safety and diagnostic support.

Purpose of the Study:

• To develop an effective method for 4D medical image interpolation.
• To enable accurate prediction of intermediate frames at arbitrary time points for continuous-time analysis.
• To address challenges of motion artifacts and blurring in low-resolution 4D medical images.

Main Methods:

• Proposed the Relation-Aware Bi-directional Motion Estimation Network (RA-BiMENet) for 4D medical image interpolation.
• Incorporated a spatiotemporal transform MLP (TS-MLP) module with a relation-aware multi-scale MLP (RAM-MLP) for nonlinear motion estimation.
• Utilized a hierarchical spatiotemporal fusion (HSTF) module with forward warping and self-attention for feature integration and detail restoration.

Main Results:

• RA-BiMENet demonstrated superior performance over state-of-the-art methods on multiple quantitative metrics.
• The network effectively generated high-fidelity and temporally coherent interpolated frames even with complex deformations.
• Validated effectiveness and superiority for continuous-time 4D medical image interpolation.

Conclusions:

• RA-BiMENet significantly advances 4D medical image interpolation by accurately predicting intermediate frames.
• The proposed method enhances the quality and temporal continuity of 4D medical images, supporting improved diagnostic capabilities.
• This approach offers a robust solution for continuous-time dynamic organ motion analysis in medical imaging.