Multi-Stage Adaptive Spline Autofocus (MASA) with a Learned Metric for Deformable Motion Compensation in...

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Cone-beam CT (CBCT) is vital in abdominal interventional imaging but suffers from motion artifacts due to long acquisition times.
• Existing image-based autofocus methods for CBCT motion compensation struggle with the high dimensionality of complex motion trajectories.

Purpose of the Study:

• To develop a novel optimization strategy, Multi-Stage Adaptive Spine Autofocus (MASA), for compensating complex deformable motion in abdominal CBCT.
• To leverage differentiable deep autofocus metrics for a more effective motion compensation approach.

Main Methods:

• MASA employs a multi-stage adaptive sampling strategy using Hermite splines to optimize motion trajectories.
• The method simultaneously optimizes sampling phase, local temporal density, and time-dependent amplitude, progressively accommodating complexity with increasing stages.
• Evaluated using simulations with sigmoid and multi-frequency motion, and validated on clinical liver CBCT data.

Main Results:

• MASA demonstrated successful motion compensation, outperforming fixed sampling strategies (0.026 vs 0.011 SSIM increase).
• The method adaptively allocated sampling density to sudden motion events without increasing dimensionality.
• A 3-stage MASA achieved a twofold SSIM increase compared to single-stage autofocus (0.076 vs 0.040).

Conclusions:

• MASA offers a novel framework for deformable motion compensation in abdominal CBCT.
• Its adaptive temporal sampling technique is key for handling complex motion trajectories.
• MASA improves delineation of both vessels and soft tissues in clinical liver CBCT datasets.