Clinical Pixel Feature Recalibration Module for Ophthalmic Image Classification

Ji Lu Zhao, Xiaoqing Zhang, Xiao Wu, Zhi Xuan Zhang, Tong Zhang, Heng Li, Yan Hu, Jiang Liu

Research output: Chapter in Book/Conference proceedingConference contributionpeer-review

Abstract

Ophthalmic image examination has become a commonly-acknowledged way for ocular disease screening and diagnosis. Clinical features extracted from ophthalmic images play different roles in affecting clinicians making diagnosis results, but how to incorporate these clinical features into convolutional neural network (CNN) representations has been less studied. In this paper, we propose a simple yet practical module, Clinical Pixel Feature Recalibration Module (CPF), aiming to exploit the potential of clinical features to improve the ocular disease recognition performance of CNNs. CPF first extracts clinical pixel features from each spatial position of all feature maps by clinical cross-channel pooling, then estimates each spatial position recalibration weight in a pixel-independent clinical fusion. By infusing the relative importance of clinical features into feature maps at the pixel level, CPF is supposed to enhance the representational ability of CNNs. Our CPF is easily inserted into existing CNNs with negligible overhead. We conduct comprehensive experiments on two publicly available ophthalmic image datasets and CIFAR datasets, and the results show the superiority and generation ability of CPF over advanced attention methods. Furthermore, this paper presents an in-depth weight visualization analysis to investigate the inherent behavior of CPF, aiming to improve the interpretability of CNNs in the decision-making process.

Original languageEnglish
Title of host publicationArtificial Neural Networks and Machine Learning – ICANN 2023 - 32nd International Conference on Artificial Neural Networks, Proceedings
EditorsLazaros Iliadis, Antonios Papaleonidas, Plamen Angelov, Chrisina Jayne
PublisherSpringer Science and Business Media Deutschland GmbH
Pages87-98
Number of pages12
ISBN (Print)9783031442155
DOIs
Publication statusPublished - 2023
Externally publishedYes
Event32nd International Conference on Artificial Neural Networks, ICANN 2023 - Heraklion, Greece
Duration: 26 Sept 202329 Sept 2023

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume14257 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference32nd International Conference on Artificial Neural Networks, ICANN 2023
Country/TerritoryGreece
CityHeraklion
Period26/09/2329/09/23

Keywords

  • CPF
  • Ophthalmic image
  • attention
  • classification
  • interpretability

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'Clinical Pixel Feature Recalibration Module for Ophthalmic Image Classification'. Together they form a unique fingerprint.

Cite this