One-Shot Neural Architecture Search with Network Similarity Directed Initialization for Pathological Image Classification

2025-06-17

AI Summary

Pathological image classification on edge devices suffers from low computational efficiency and poor generalization due to the misalignment between generic vision models and pathological domain characteristics—such as staining variability and multi-scale semantic structures. Method: We propose a one-shot neural architecture search (NAS) framework integrating network similarity-guided initialization (NSDI) and domain adaptation. NSDI enhances NAS convergence stability by leveraging architectural similarity to pre-trained models; domain adaptation is embedded into the one-shot search process for joint optimization of architecture discovery and cross-domain feature alignment. Pathology-specific regularization and differentiable architecture optimization further tailor the search to clinical requirements. Results: On the BRACS dataset, our method achieves state-of-the-art classification accuracy, improved clinical interpretability via precise lesion localization, and significantly reduced inference latency and resource consumption on edge devices.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Deep Neural Architectures and Foundation Models
• Constraint Satisfaction and Optimization: Search

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Deep learning-based pathological image analysis presents unique challenges due to the practical constraints of network design. Most existing methods apply computer vision models directly to medical tasks, neglecting the distinct characteristics of pathological images. This mismatch often leads to computational inefficiencies, particularly in edge-computing scenarios. To address this, we propose a novel Network Similarity Directed Initialization (NSDI) strategy to improve the stability of neural architecture search (NAS). Furthermore, we introduce domain adaptation into one-shot NAS to better handle variations in staining and semantic scale across pathology datasets. Experiments on the BRACS dataset demonstrate that our method outperforms existing approaches, delivering both superior classification performance and clinically relevant feature localization.

Problem

• Improves neural architecture search stability for pathological images
• Addresses staining and scale variations in pathology datasets
• Enhances classification and feature localization in medical imaging

Innovation

• Network Similarity Directed Initialization strategy
• Domain adaptation in one-shot NAS
• Improved classification and feature localization