Multi-level Cellular Automata for FLIM networks

TL;DR

This work tackles Salient Object Detection in data-scarce medical imaging by fusing Feature Learning from Image Markers (FLIM) with Cellular Automata (CA). A multi-level CA framework initializes separate CA states from saliency maps decoded at each FLIM encoder layer, forming an ensemble that is evolved and then merged into a final saliency map. The approach achieves competitive performance against lightweight and deep SOD models on brain tumor MRI slices and parasite egg detection while using far fewer parameters and only weak annotations for training. The results demonstrate the practicality of FLIM-driven CA initialization for robust, low-resource saliency in heterogeneous medical imaging domains, with potential for real-time deployment and further extensions to 3D multi-label segmentation.

Abstract

The necessity of abundant annotated data and complex network architectures presents a significant challenge in deep-learning Salient Object Detection (deep SOD) and across the broader deep-learning landscape. This challenge is particularly acute in medical applications in developing countries with limited computational resources. Combining modern and classical techniques offers a path to maintaining competitive performance while enabling practical applications. Feature Learning from Image Markers (FLIM) methodology empowers experts to design convolutional encoders through user-drawn markers, with filters learned directly from these annotations. Recent findings demonstrate that coupling a FLIM encoder with an adaptive decoder creates a flyweight network suitable for SOD, requiring significantly fewer parameters than lightweight models and eliminating the need for backpropagation. Cellular Automata (CA) methods have proven successful in data-scarce scenarios but require proper initialization -- typically through user input, priors, or randomness. We propose a practical intersection of these approaches: using FLIM networks to initialize CA states with expert knowledge without requiring user interaction for each image. By decoding features from each level of a FLIM network, we can initialize multiple CAs simultaneously, creating a multi-level framework. Our method leverages the hierarchical knowledge encoded across different network layers, merging multiple saliency maps into a high-quality final output that functions as a CA ensemble. Benchmarks across two challenging medical datasets demonstrate the competitiveness of our multi-level CA approach compared to established models in the deep SOD literature.

Figures (15)

• Figure 1: Multi-Level Cellular Automata for FLIM networks: (1) Users iteratively select training images and define FLIM-Encoder architecture. (2) During inference, images enter the encoder, and features decode into intermediary saliency maps, initializing CA cells. (3) Multi-level saliency maps evolve from this initialization, and (4) merge into a final saliency.
• Figure 2: Adaptive decoder for FLIM-Encoder layer 3 (same across all $l$ layers): (a) input with foreground/background markers; (b) channel $i$ with $w_i = 1$ (foreground); (c) channel $j$ with $w_j = -1$ (background); (d) resulting saliency map $\mathcal{S}_3$.
• Figure 3: Brain image as a CA's lattice of cells, showing individual cells and their neighboring relationships.
• Figure 4: Multi-Level Cellular Automata for FLIM networks (example for a 4-layer FLIM Network): (1) Users iteratively select training images and define FLIM-Encoder architecture. (2) During inference, images enter the encoder, and features decode into intermediary saliency maps, initializing CA cells. (3) Multi-level saliency maps evolve from this initialization, and (4) merge into a final saliency.
• Figure 5: Examples of user annotations: red markers indicate tumors or parasite eggs; white markers show healthy brain tissue, background, or debris.