Hidden Layer Visualization for Convolutional Neural Networks: A Brief Review

Tasks in the field of computer vision are mostly led by convolutional neural networks (CNNs) (Aamir et al. in Electronics 11(1), 2022 [1]), however, understanding and interpreting the information within these networks remains a challenge. To gain a deeper understanding of how a network learns and functions, it is imperative to develop visualization tools to address these complex structures. This area remains a crucial point of research to advance the understanding of deep neural network operations. Therefore, this paper presents a comprehensive review aimed at establishing the fundamental framework of the methodologies employed in the visualization of hidden layers in CNNs. Approaches such as activation maximization, hidden layer feature analysis, and post hoc visualization techniques are specifically addressed. The focus is on the application of CNN in cancer diagnostics, evaluating the feasibility and utility of hidden layer visualization methodologies in this context. As a future perspective, research and development of a layered visualization model that optimizes the performance of neural networks in medical image analysis is proposed.