Multi-stages attention Breast cancer classification based on nonlinear spiking neural P neurons with autapses

TL;DR

This work tackles breast cancer histopathology classification on BreakHis by addressing information loss from downsampling through a multi-stages attention framework and a memory-enabled NSNP neuron with autapses. The method combines a ShuffleNet backbone with stage-wise spatial attention, channel attention, and an NSNP module to encode and progressively reduce feature representations for robust classification. Empirical results show a 8-class accuracy of 96.32% across magnifications and strong binary-class performance, with ablation confirming significant contributions from both the multi-stages attention and NSNP components. The approach demonstrates competitive performance against recent methods and introduces memory-augmented, attention-guided feature processing that is relevant for medical image analysis.

Abstract

Breast cancer(BC) is a prevalent type of malignant tumor in women. Early diagnosis and treatment are vital for enhancing the patients' survival rate. Downsampling in deep networks may lead to loss of information, so for compensating the detail and edge information and allowing convolutional neural networks to pay more attention to seek the lesion region, we propose a multi-stages attention architecture based on NSNP neurons with autapses. First, unlike the single-scale attention acquisition methods of existing methods, we set up spatial attention acquisition at each feature map scale of the convolutional network to obtain an fusion global information on attention guidance. Then we introduce a new type of NSNP variants called NSNP neurons with autapses. Specifically, NSNP systems are modularized as feature encoders, recoding the features extracted from convolutional neural network as well as the fusion of attention information and preserve the key characteristic elements in feature maps. This ensures the retention of valuable data while gradually transforming high-dimensional complicated info into low-dimensional ones. The proposed method is evaluated on the public dataset BreakHis at various magnifications and classification tasks. It achieves a classification accuracy of 96.32% at all magnification cases, outperforming state-of-the-art methods. Ablation studies are also performed, verifying the proposed model's efficacy. The source code is available at XhuBobYoung/Breast-cancer-Classification.

Figures (9)

• Figure 1: Attention heat map at different stages. As the network deepens, the downsampling operation makes the image lower in resolution, while losing quite a lot of detail information. Stage 2 and stage 3 show less detail than stage 1. when all the attention maps are finally fused, the attention image shows a completely different state.
• Figure 2: An NSNP neuron with autapses. $\sigma_1,\dots, \sigma_m$ represent m predecessor neurons, and $w_1,\dots, w_m$ represent the corresponding weight. $z^{-1}$ denotes an adaptive transformation of the state $u_i(t)$.
• Figure 3: The architecture of the proposed network. SA and CA denote spatial and channel attention modules.
• Figure 4: NSNP module. F1-F4 are feature maps with four dimensions. All A1, A2, A3, and O1 are vectors with $C$ elements.
• Figure 5: Data enhancement techniques of different types. (a) Original image;(b) Rotation; (c) Histogram equalization and rotation; (d) Solarization and automatic contrast enhancement.