Equivariant Filters for Efficient Tracking in 3D Imaging

TL;DR

This paper introduces an efficient 3D object tracking method using equivariant filters that maintain transformation properties, enabling real-time performance with fixed computational cost in medical imaging applications.

Contribution

The paper presents a novel architecture that replaces traditional convolutional and fully connected layers with equivariant filters, improving efficiency and transformation preservation in 3D tracking.

Findings

Achieves state-of-the-art tracking accuracy

Maintains fixed computational cost regardless of input complexity

Effective in real-time fetal and adult brain MRI tracking

Abstract

We demonstrate an object tracking method for 3D images with fixed computational cost and state-of-the-art performance. Previous methods predicted transformation parameters from convolutional layers. We instead propose an architecture that does not include either flattening of convolutional features or fully connected layers, but instead relies on equivariant filters to preserve transformations between inputs and outputs (e.g. rot./trans. of inputs rotate/translate outputs). The transformation is then derived in closed form from the outputs of the filters. This method is useful for applications requiring low latency, such as real-time tracking. We demonstrate our model on synthetically augmented adult brain MRI, as well as fetal brain MRI, which is the intended use-case.