Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Longitudinal Lesion Inpainting in Brain MRI via 3D Region Aware Diffusion

Zahra Karimaghaloo, Dumitru Fetco, Haz-Edine Assemlal, Hassan Rivaz, Douglas L. Arnold

TL;DR

A novel pseudo-3D longitudinal inpainting framework based on Denoising Diffusion Probabilistic Models (DDPM), which utilizes multi-channel conditioning to incorporate longitudinal context from distinct visits and extends Region-Aware Diffusion to the medical domain, focusing the generative process on pathological regions without altering surrounding healthy tissue.

Abstract

Accurate longitudinal analysis of brain MRI is often hindered by evolving lesions, which bias automated neuroimaging pipelines. While deep generative models have shown promise in inpainting these lesions, most existing methods operate cross-sectionally or lack 3D anatomical continuity. We present a novel pseudo-3D longitudinal inpainting framework based on Denoising Diffusion Probabilistic Models (DDPM). Our approach utilizes multi-channel conditioning to incorporate longitudinal context from distinct visits (t_1, t_2) and extends Region-Aware Diffusion (RAD) to the medical domain, focusing the generative process on pathological regions without altering surrounding healthy tissue. We evaluated our model against state-of-the-art baselines on longitudinal brain MRI from 93 patients. Our model significantly outperforms the leading baseline (FastSurfer-LIT) in terms of perceptual fidelity, reducing the Learned Perceptual Image Patch Similarity (LPIPS) distance from 0.07 to 0.03 while effectively eliminating inter-slice discontinuities. Furthermore, our model demonstrates high longitudinal stability with a Temporal Fidelity Index of 1.024, closely approaching the ideal value of 1.0 and substantially narrowing the gap compared to LIT's TFI of 1.22. Notably, the RAD mechanism provides a substantial gain in efficiency; our framework achieves an average processing time of 2.53 min per volume, representing approximately 10x speedup over the 24.30 min required by LIT. By leveraging longitudinal priors and region-specific denoising, our framework provides a highly reliable and efficient preprocessing step for the study of progressive neurodegenerative diseases. A derivative dataset consisting of 93 pre-processed scans used for testing will be available upon request after acceptance. Code will be released upon acceptance.

Longitudinal Lesion Inpainting in Brain MRI via 3D Region Aware Diffusion

TL;DR

A novel pseudo-3D longitudinal inpainting framework based on Denoising Diffusion Probabilistic Models (DDPM), which utilizes multi-channel conditioning to incorporate longitudinal context from distinct visits and extends Region-Aware Diffusion to the medical domain, focusing the generative process on pathological regions without altering surrounding healthy tissue.

Abstract

Accurate longitudinal analysis of brain MRI is often hindered by evolving lesions, which bias automated neuroimaging pipelines. While deep generative models have shown promise in inpainting these lesions, most existing methods operate cross-sectionally or lack 3D anatomical continuity. We present a novel pseudo-3D longitudinal inpainting framework based on Denoising Diffusion Probabilistic Models (DDPM). Our approach utilizes multi-channel conditioning to incorporate longitudinal context from distinct visits (t_1, t_2) and extends Region-Aware Diffusion (RAD) to the medical domain, focusing the generative process on pathological regions without altering surrounding healthy tissue. We evaluated our model against state-of-the-art baselines on longitudinal brain MRI from 93 patients. Our model significantly outperforms the leading baseline (FastSurfer-LIT) in terms of perceptual fidelity, reducing the Learned Perceptual Image Patch Similarity (LPIPS) distance from 0.07 to 0.03 while effectively eliminating inter-slice discontinuities. Furthermore, our model demonstrates high longitudinal stability with a Temporal Fidelity Index of 1.024, closely approaching the ideal value of 1.0 and substantially narrowing the gap compared to LIT's TFI of 1.22. Notably, the RAD mechanism provides a substantial gain in efficiency; our framework achieves an average processing time of 2.53 min per volume, representing approximately 10x speedup over the 24.30 min required by LIT. By leveraging longitudinal priors and region-specific denoising, our framework provides a highly reliable and efficient preprocessing step for the study of progressive neurodegenerative diseases. A derivative dataset consisting of 93 pre-processed scans used for testing will be available upon request after acceptance. Code will be released upon acceptance.
Paper Structure (4 sections, 5 figures, 2 tables)

This paper contains 4 sections, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Experiments
  4. Conclusions

Figures (5)

  • Figure 1: P3D-RAD Architecture. The model integrates Pseudo-3D convolutions with a Region A Diffusion (RAD) mechanism. Spatial FiLM parameters are injected into both the encoder and decoder P3D ResBlocks (blue lines) to provide temporally and spatially aware feature modulation.
  • Figure 2: Qualitative comparison of 3D inpainting. Column 1 shows the ground truth; subsequent columns show model results (Rows 1–3) and their absolute error maps (Rows 5–7). Row 4 displays zoomed-in versions of the regions marked with light green within the sagittal view (Row 2). Dark green outlines the inpaiting mask. As observed our inpainting results shows the highest fidelity to the ground truth.
  • Figure 3: Longitudinal Inpainting. Inpainted $t_2$ across four models. Row 1: original at $t_1$ with no inpainting (same for all); Row 2: original and inpainted at $t_2$; Row 3: Absolute difference maps. Dark green outlines the inpaiting mask. LIT appears blurred due to global blurring in its pipeline.
  • Figure 4: LPIPS-based longitudinal change preservation. Scatter compares original vs inpainted LPIPS between $(t_1,t_2)$. Red: linear fit; dashed black: identity ($y=x$). Our method achieves the highest Pearson ($r$) and Temporal Fidelity Index closest to 1.
  • Figure 5: Expert Blind Review. Two examples of real MS lesion inpainting (green outlines). Blue arrows highlight structural artifacts in the LIT baseline identified by the expert.