Joint Learning with Local and Global Consistency for Improved Medical Image Segmentation

Md Atik Ahamed, Abdullah Al Zubaer Imran

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

2 Scopus citations


Semantic segmentation has been one of the key components in subsequent image-based decision-making across computer vision and biomedical imaging. While a lot of progress has been made with the advent of deep learning, segmentation models rely heavily on large labeled datasets for optimal results. Moreover, with added challenges due to varying imaging conditions, abnormalities, etc., it becomes relatively a harder problem to solve, even by the most sophisticated models. Additionally, segmentation models when employed at small patch-level lose the global context and when employed at the full image-level may lose focus to closely located and small objects-of-interest. In order to resolve such issues and thereby improve the segmentation performance, we propose a novel joint patch- and image-level training framework namely Image-to-Patch w/ Patch-to-Image (IPPI) which at the same time preserves the global context and pays attention to local details. Accommodating the joint training, our proposed IPPI technique can be incorporated with any segmentation network for improved performance and local-global consistency. Our experimentation with three different segmentation networks (U-Net, U-Net++, and NodeU-Net) in segmenting cell nuclei and retinal vessel demonstrates the effectiveness of the proposed IPPI method. The segmentation improvements—13.35% over U-Net, 5.56% over U-Net++, and 4.59% over NodeU-Net IoU (Intersection over Union) make it a potentially beneficial tool in challenging segmentation tasks.

Original languageEnglish
Title of host publicationMedical Image Understanding and Analysis - 26th Annual Conference, MIUA 2022, Proceedings
EditorsGuang Yang, Angelica Aviles-Rivero, Michael Roberts, Carola-Bibiane Schönlieb
Number of pages15
StatePublished - 2022
Event26th Annual Conference on Medical Image Understanding and Analysis, MIUA 2022 - Cambridge, United Kingdom
Duration: Jul 27 2022Jul 29 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13413 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Conference26th Annual Conference on Medical Image Understanding and Analysis, MIUA 2022
Country/TerritoryUnited Kingdom

Bibliographical note

Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.


  • Biomedical imaging
  • CNN
  • Computer vision
  • Deep learning
  • Semantic segmentation
  • U-Net

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science


Dive into the research topics of 'Joint Learning with Local and Global Consistency for Improved Medical Image Segmentation'. Together they form a unique fingerprint.

Cite this