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

Marawan Elbatel,Konstantinos Kamnitsas,Xiaomeng Li

2024-07-03

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: <a class="link-external link-https" href="https://github.com/xmed-lab/GeCA" rel="external noopener nofollow">this https URL</a>.

Computer Vision and Pattern Recognition,Artificial Intelligence

What problem does this paper attempt to address?

This paper proposes a new model called Generative Cellular Automaton (GeCA), which is inspired by the biological evolution process, especially the process of developing from a single cell into an organism. Current generative models, such as Generative Adversarial Networks (GANs) and Denoising Diffusion Probabilistic Models (DDPMs), approach real data in terms of statistical attributes to generate new data, but the perspective of biological evolution has not been fully explored. In medical image analysis, especially in the diagnosis of retinal diseases, data scarcity and class imbalance pose a challenge. In the paper, the GeCA model is evaluated as an effective enhancement tool for retinal disease classification, particularly in the field of Optical Coherence Tomography (OCT) images where data is scarce. By using GeCA to augment the OCT dataset, the average F1 score of 11 different eye diseases improved by 12%, outperforming traditional baselines and diffusion methods based on UNet or Transformers. The GeCA model combines Neural Cellular Automata (NCA) and diffusion targets, allowing for high-resolution image synthesis while maintaining fewer parameters. It introduces a method called Gene Genetic Guidance (GHG), which improves the image sampling process, surpassing the state-of-the-art Diffusion Transformer (DiT) model in image generation and disease classification tasks, while only having half the parameters of DiT. Overall, the paper addresses the problem of utilizing concepts from biological evolution to improve generative models, especially in the context of medical image analysis to tackle the challenges posed by data scarcity and class imbalance. The GeCA model demonstrates potential in generating high-resolution images and improving disease recognition performance.

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

OA-GAN: Organ-Aware Generative Adversarial Network for Synthesizing Contrast-Enhanced Medical Images

A generative adversarial network with multi-scale convolution and dilated convolution res-network for OCT retinal image despeckling

Cellcyclegan: Spatiotemporal Microscopy Image Synthesis Of Cell Populations Using Statistical Shape Models And Conditional Gans

Assessment of Generative Adversarial Networks Model for Synthetic Optical Coherence Tomography Images of Retinal Disorders

Neural Cell Video Synthesis via Optical-Flow Diffusion

Diffusion Models-based Data Augmentation for the Cell Cycle Phase Classification

Transfer Learning and Interpretable Analysis-Based Quality Assessment of Synthetic Optical Coherence Tomography Images by CGAN Model for Retinal Diseases

High-resolution medical image synthesis using progressively grown generative adversarial networks

Synthesizing New Retinal Symptom Images by Multiple Generative Models

Generating OCT B-Scan DME images using optimized Generative Adversarial Networks (GANs)

Synthesizing Retinal and Neuronal Images with Generative Adversarial Nets

Diffusion-based data augmentation methodology for improved performance in ocular disease diagnosis using retinography images

Generative Adversarial Networks Synthesize Realistic OCT Images of the Retina

Retinal Image Synthesis from Multiple-Landmarks Input with Generative Adversarial Networks.

Generation of Retinal OCT Images with Diseases Based on Cgan

Super-Resolution Reconstruction of Cell Images Based on Generative Adversarial Networks

Frequency-Time Diffusion with Neural Cellular Automata

Cascaded Diffusion Models for 2D and 3D Microscopy Image Synthesis to Enhance Cell Segmentation

Abstract We016: Generative Artificial Intelligence for hPSC-derived Cardiac Organoid Florescence Generation