Biologically-inspired Semi-supervised Semantic Segmentation for Biomedical Imaging

2024-12-04

arXiv.org

AI Summary

To address the scarcity of annotated data in biomedical image segmentation, this paper proposes a two-stage semi-supervised framework. In the first stage, a biologically inspired, backpropagation-free unsupervised pretraining is performed on the entire encoder-decoder architecture—including both convolutional and transposed convolutional layers—guided by Hebbian learning principles (“neurons that fire together, wire together”). In the second stage, the pretrained model is fine-tuned using only a small number of labeled samples. This work is the first to systematically integrate Hebbian-style local plasticity mechanisms into semantic segmentation architectures, enabling end-to-end, gradient-free feature initialization. Extensive experiments on multiple biomedical benchmarks demonstrate substantial improvements over state-of-the-art methods: the approach achieves full-supervision-level performance using merely 10% of labeled data. Moreover, its unsupervised initialization is model-agnostic and transferable, effectively enhancing few-shot generalization across mainstream segmentation networks.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Semi-Supervised Learning
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We propose a novel two-stage semi-supervised learning approach for training downsampling-upsampling semantic segmentation architectures. The first stage does not use backpropagation. Rather, it exploits the bio-inspired Hebbian principle"fire together, wire together"as a local learning rule for updating the weights of both convolutional and transpose-convolutional layers, allowing unsupervised discovery of data features. In the second stage, the model is fine-tuned with standard backpropagation on a small subset of labeled data. We evaluate our methodology through experiments conducted on several widely used biomedical datasets, deeming that this domain is paramount in computer vision and is notably impacted by data scarcity. Results show that our proposed method outperforms SOTA approaches across different levels of label availability. Furthermore, we show that using our unsupervised stage to initialize the SOTA approaches leads to performance improvements. The code to replicate our experiments can be found at: https://github.com/ciampluca/hebbian-medical-image-segmentation

Problem

• Develops a bio-inspired semi-supervised learning method for semantic segmentation.
• Addresses data scarcity in biomedical imaging using unsupervised feature discovery.
• Improves performance of SOTA approaches with unsupervised initialization.

Innovation

• Hebbian principle for unsupervised feature discovery
• Semi-supervised learning with labeled data fine-tuning
• Outperforms SOTA in biomedical image segmentation