Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Adversarial Approach to Register Extreme Resolution Tissue Cleared 3D Brain Images

Abdullah Naziba, Clinton Fookes, Dimitri Perrin

TL;DR

The paper tackles registering extremely high-resolution tissue-cleared brain images, where traditional optimization-based methods are impractical. It introduces InvGAN, an inverse-consistent patch-based registration framework with two decoders and two discriminators $D_S$ and $D_T$, trained using a combination of cross-correlation similarity, cycle consistency, and adversarial losses. Across 25% and 100% CUBIC/LSPM datasets, InvGAN achieves state-of-the-art or near state-of-the-art registration accuracy while drastically reducing computation time (minutes rather than hours) and mitigating patch artifacts through adversarial training. Landmark validation confirms strong local alignment, and successful 100% resolution experiments demonstrate feasibility on ultra-large volumes with modest hardware, offering a scalable reference for future high-resolution brain image analysis pipelines.

Abstract

We developed a generative patch based 3D image registration model that can register very high resolution images obtained from a biochemical process name tissue clearing. Tissue clearing process removes lipids and fats from the tissue and make the tissue transparent. When cleared tissues are imaged with Light-sheet fluorescent microscopy, the resulting images give a clear window to the cellular activities and dynamics inside the tissue.Thus the images obtained are very rich with cellular information and hence their resolution is extremely high (eg .2560x2160x676). Analyzing images with such high resolution is a difficult task for any image analysis pipeline.Image registration is a common step in image analysis pipeline when comparison between images are required. Traditional image registration methods fail to register images with such extant. In this paper we addressed this very high resolution image registration issue by proposing a patch-based generative network named InvGAN. Our proposed network can register very high resolution tissue cleared images. The tissue cleared dataset used in this paper are obtained from a tissue clearing protocol named CUBIC. We compared our method both with traditional and deep-learning based registration methods.Two different versions of CUBIC dataset are used, representing two different resolutions 25% and 100% respectively. Experiments on two different resolutions clearly show the impact of resolution on the registration quality. At 25% resolution, our method achieves comparable registration accuracy with very short time (7 minutes approximately). At 100% resolution, most of the traditional registration methods fail except Elastix registration tool.Elastix takes 28 hours to register where proposed InvGAN takes only 10 minutes.

An Adversarial Approach to Register Extreme Resolution Tissue Cleared 3D Brain Images

TL;DR

The paper tackles registering extremely high-resolution tissue-cleared brain images, where traditional optimization-based methods are impractical. It introduces InvGAN, an inverse-consistent patch-based registration framework with two decoders and two discriminators and , trained using a combination of cross-correlation similarity, cycle consistency, and adversarial losses. Across 25% and 100% CUBIC/LSPM datasets, InvGAN achieves state-of-the-art or near state-of-the-art registration accuracy while drastically reducing computation time (minutes rather than hours) and mitigating patch artifacts through adversarial training. Landmark validation confirms strong local alignment, and successful 100% resolution experiments demonstrate feasibility on ultra-large volumes with modest hardware, offering a scalable reference for future high-resolution brain image analysis pipelines.

Abstract

We developed a generative patch based 3D image registration model that can register very high resolution images obtained from a biochemical process name tissue clearing. Tissue clearing process removes lipids and fats from the tissue and make the tissue transparent. When cleared tissues are imaged with Light-sheet fluorescent microscopy, the resulting images give a clear window to the cellular activities and dynamics inside the tissue.Thus the images obtained are very rich with cellular information and hence their resolution is extremely high (eg .2560x2160x676). Analyzing images with such high resolution is a difficult task for any image analysis pipeline.Image registration is a common step in image analysis pipeline when comparison between images are required. Traditional image registration methods fail to register images with such extant. In this paper we addressed this very high resolution image registration issue by proposing a patch-based generative network named InvGAN. Our proposed network can register very high resolution tissue cleared images. The tissue cleared dataset used in this paper are obtained from a tissue clearing protocol named CUBIC. We compared our method both with traditional and deep-learning based registration methods.Two different versions of CUBIC dataset are used, representing two different resolutions 25% and 100% respectively. Experiments on two different resolutions clearly show the impact of resolution on the registration quality. At 25% resolution, our method achieves comparable registration accuracy with very short time (7 minutes approximately). At 100% resolution, most of the traditional registration methods fail except Elastix registration tool.Elastix takes 28 hours to register where proposed InvGAN takes only 10 minutes.

Paper Structure

This paper contains 17 sections, 8 equations, 29 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Materials and Method
  4. Network Architecture
  5. Discriminator Loss
  6. Generator Loss
  7. Data Preprocessing
  8. Training Patch Selection
  9. Training
  10. Competing Methods
  11. Results
  12. Results at 25% Resolution
  13. Results at 100% Resolution
  14. Landmark Validation
  15. Computation Time
  16. ...and 2 more sections

Figures (29)

  • Figure 1: Inverse Consistent Adversarial Network with two discriminators $D_S$ and $D_T$, where $D_S$ compares the target image patch with the transformed source patch, and $D_T$ compares the source image patch with the transformed target patch.
  • Figure 2: no adversarial loss
  • Figure 3: with adversarial loss
  • Figure 4: Box artifacts in non-Adversarial training of InvGAN
  • Figure 5: Visual Comparison at 25% Resolution-1
  • ...and 24 more figures