Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Online learning in motion modeling for intra-interventional image sequences

Niklas Gunnarsson, Jens Sjölund, Peter Kimstrand, Thomas. B Schön

TL;DR

A probabilistic motion model for sequential medical images, with the ability to both estimate motion between acquired images and forecast the motion ahead of time, is presented, showing reliable motion estimates and an improved forecasting performance using patient-specific adaptation by online learning.

Abstract

Image monitoring and guidance during medical examinations can aid both diagnosis and treatment. However, the sampling frequency is often too low, which creates a need to estimate the missing images. We present a probabilistic motion model for sequential medical images, with the ability to both estimate motion between acquired images and forecast the motion ahead of time. The core is a low-dimensional temporal process based on a linear Gaussian state-space model with analytically tractable solutions for forecasting, simulation, and imputation of missing samples. The results, from two experiments on publicly available cardiac datasets, show reliable motion estimates and an improved forecasting performance using patient-specific adaptation by online learning.

Online learning in motion modeling for intra-interventional image sequences

TL;DR

A probabilistic motion model for sequential medical images, with the ability to both estimate motion between acquired images and forecast the motion ahead of time, is presented, showing reliable motion estimates and an improved forecasting performance using patient-specific adaptation by online learning.

Abstract

Image monitoring and guidance during medical examinations can aid both diagnosis and treatment. However, the sampling frequency is often too low, which creates a need to estimate the missing images. We present a probabilistic motion model for sequential medical images, with the ability to both estimate motion between acquired images and forecast the motion ahead of time. The core is a low-dimensional temporal process based on a linear Gaussian state-space model with analytically tractable solutions for forecasting, simulation, and imputation of missing samples. The results, from two experiments on publicly available cardiac datasets, show reliable motion estimates and an improved forecasting performance using patient-specific adaptation by online learning.

Paper Structure

This paper contains 10 sections, 13 equations, 4 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Background - Linear Gaussian state space model
  4. Method
  5. Online learning
  6. Implementation details
  7. Experiments
  8. Discussion and conclusion
  9. Acknowledgments.
  10. Disclosure of Interests.

Figures (4)

  • Figure 1: An observation $y_t$ is downsampled to the temporal process \ref{['fig:graph_i']}. The spatial transformation is generated given $y_0$ and the low-dimensional motion model \ref{['fig:graph_g']}. \ref{['fig:graph']} visualizes the entire model.
  • Figure 2: During online-learning we fix the parameters of the encoder ($\phi$) and the decoder ($\theta$), and only update the parameters of the LG-SSM ($\gamma$).
  • Figure 3: Overlay of true sequence (magenta), and $\varphi_t = 0$, on top, and our estimation given every 10th sample, on bottom (green). On the right, the distribution of the left ventricle area for $20$ latent samples under three scenarios: all time points observed, every 5th, and every 10th. The figure is colored in the online version.
  • Figure 4: Overlay between true sequence (magenta) and forecasted sequence (green) using pre-trained model, on top, and online learning, on bottom. To the right, Dice score distribution of the left ventricle from 20 forecasted samples and the estimated region given the entire sequence. The figure is colored in the online version.