Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Organism Starts with a Single Pix-Cell: A Neural Cellular Diffusion for High-Resolution Image Synthesis

Marawan Elbatel, Konstantinos Kamnitsas, Xiaomeng Li

TL;DR

The paper addresses data scarcity and class imbalance in OCT-based retinal disease classification by proposing Generative Cellular Automata (GeCA), a diffusion-guided Neural Cellular Automata framework that evolves pix-cells to synthesize high-resolution Fundus and OCT images. It introduces Gene Heredity Guidance (GHG) to improve reverse sampling, and employs a latent-space diffusion objective with a single Diffusion Transformer block and localized attention to maintain both local realism and global coherence. GeCA demonstrates superior image quality and downstream performance, surpassing state-of-the-art diffusion transformers while using fewer parameters, and, when used to augment OCT datasets, yields a substantial improvement in multi-label classification metrics (e.g., up to a 12% average F1 gain). The work highlights a data-efficient generative approach with potential applicability beyond ophthalmology, and suggests future directions such as update scheduling and cross-domain validation.

Abstract

Generative modeling seeks to approximate the statistical properties of real data, enabling synthesis of new data that closely resembles the original distribution. Generative Adversarial Networks (GANs) and Denoising Diffusion Probabilistic Models (DDPMs) represent significant advancements in generative modeling, drawing inspiration from game theory and thermodynamics, respectively. Nevertheless, the exploration of generative modeling through the lens of biological evolution remains largely untapped. In this paper, we introduce a novel family of models termed Generative Cellular Automata (GeCA), inspired by the evolution of an organism from a single cell. GeCAs are evaluated as an effective augmentation tool for retinal disease classification across two imaging modalities: Fundus and Optical Coherence Tomography (OCT). In the context of OCT imaging, where data is scarce and the distribution of classes is inherently skewed, GeCA significantly boosts the performance of 11 different ophthalmological conditions, achieving a 12% increase in the average F1 score compared to conventional baselines. GeCAs outperform both diffusion methods that incorporate UNet or state-of-the art variants with transformer-based denoising models, under similar parameter constraints. Code is available at: https://github.com/xmed-lab/GeCA.

An Organism Starts with a Single Pix-Cell: A Neural Cellular Diffusion for High-Resolution Image Synthesis

TL;DR

The paper addresses data scarcity and class imbalance in OCT-based retinal disease classification by proposing Generative Cellular Automata (GeCA), a diffusion-guided Neural Cellular Automata framework that evolves pix-cells to synthesize high-resolution Fundus and OCT images. It introduces Gene Heredity Guidance (GHG) to improve reverse sampling, and employs a latent-space diffusion objective with a single Diffusion Transformer block and localized attention to maintain both local realism and global coherence. GeCA demonstrates superior image quality and downstream performance, surpassing state-of-the-art diffusion transformers while using fewer parameters, and, when used to augment OCT datasets, yields a substantial improvement in multi-label classification metrics (e.g., up to a 12% average F1 gain). The work highlights a data-efficient generative approach with potential applicability beyond ophthalmology, and suggests future directions such as update scheduling and cross-domain validation.

Abstract

Generative modeling seeks to approximate the statistical properties of real data, enabling synthesis of new data that closely resembles the original distribution. Generative Adversarial Networks (GANs) and Denoising Diffusion Probabilistic Models (DDPMs) represent significant advancements in generative modeling, drawing inspiration from game theory and thermodynamics, respectively. Nevertheless, the exploration of generative modeling through the lens of biological evolution remains largely untapped. In this paper, we introduce a novel family of models termed Generative Cellular Automata (GeCA), inspired by the evolution of an organism from a single cell. GeCAs are evaluated as an effective augmentation tool for retinal disease classification across two imaging modalities: Fundus and Optical Coherence Tomography (OCT). In the context of OCT imaging, where data is scarce and the distribution of classes is inherently skewed, GeCA significantly boosts the performance of 11 different ophthalmological conditions, achieving a 12% increase in the average F1 score compared to conventional baselines. GeCAs outperform both diffusion methods that incorporate UNet or state-of-the art variants with transformer-based denoising models, under similar parameter constraints. Code is available at: https://github.com/xmed-lab/GeCA.
Paper Structure (10 sections, 8 equations, 7 figures, 3 tables)

This paper contains 10 sections, 8 equations, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Generative Cellular Automata
  3. An Organism Starts With a Single Pix-Cell
  4. Cellular Diffusion: Evolving Cells into Organisms
  5. Improved Reverse Sampling via Gene Heredity
  6. Retinal Disease Classification
  7. Experiments
  8. Conclusion
  9. Acknowledgments.
  10. Disclosure of Interests.

Figures (7)

  • Figure 1: Selected synthetic images from our GeCA trained on Fundus and OCT.
  • Figure 2: GeCA overall framework.
  • Figure 3: Results summary on a public fundus dataset.
  • Figure 4: Qualitative examples for the Fundus and OCT datasets are provided; images are downsampled for visualization purposes, with high-resolution versions available in the supplementary material.
  • Figure 5: Ablation of the proposed Gene Heredity Guidance (GHG).
  • ...and 2 more figures