Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explainable AI (XAI) in Image Segmentation in Medicine, Industry, and Beyond: A Survey

Rokas Gipiškis, Chun-Wei Tsai, Olga Kurasova

TL;DR

This survey tackles the gap in explainable AI for semantic image segmentation by cataloging methods tailored to dense predictions across medicine, industry, and beyond. It introduces a method-centered taxonomy with five families—prototype-based, gradient-based, perturbation-based, counterfactual, and architecture-based—while mapping these approaches to practical datasets and evaluation metrics. The authors analyze limitations of current XAI techniques, propose benchmarks and discussion points for safety, robustness, and ground-truth explanations, and highlight the growing role of self-supervised and weakly supervised segmentation. The work underscores the practical importance of interpretable segmentation in high-stakes domains and outlines concrete future directions to advance reliable, explainable dense predictions.

Abstract

Artificial Intelligence (XAI) has found numerous applications in computer vision. While image classification-based explainability techniques have garnered significant attention, their counterparts in semantic segmentation have been relatively neglected. Given the prevalent use of image segmentation, ranging from medical to industrial deployments, these techniques warrant a systematic look. In this paper, we present the first comprehensive survey on XAI in semantic image segmentation. This work focuses on techniques that were either specifically introduced for dense prediction tasks or were extended for them by modifying existing methods in classification. We analyze and categorize the literature based on application categories and domains, as well as the evaluation metrics and datasets used. We also propose a taxonomy for interpretable semantic segmentation, and discuss potential challenges and future research directions.

Explainable AI (XAI) in Image Segmentation in Medicine, Industry, and Beyond: A Survey

TL;DR

This survey tackles the gap in explainable AI for semantic image segmentation by cataloging methods tailored to dense predictions across medicine, industry, and beyond. It introduces a method-centered taxonomy with five families—prototype-based, gradient-based, perturbation-based, counterfactual, and architecture-based—while mapping these approaches to practical datasets and evaluation metrics. The authors analyze limitations of current XAI techniques, propose benchmarks and discussion points for safety, robustness, and ground-truth explanations, and highlight the growing role of self-supervised and weakly supervised segmentation. The work underscores the practical importance of interpretable segmentation in high-stakes domains and outlines concrete future directions to advance reliable, explainable dense predictions.

Abstract

Artificial Intelligence (XAI) has found numerous applications in computer vision. While image classification-based explainability techniques have garnered significant attention, their counterparts in semantic segmentation have been relatively neglected. Given the prevalent use of image segmentation, ranging from medical to industrial deployments, these techniques warrant a systematic look. In this paper, we present the first comprehensive survey on XAI in semantic image segmentation. This work focuses on techniques that were either specifically introduced for dense prediction tasks or were extended for them by modifying existing methods in classification. We analyze and categorize the literature based on application categories and domains, as well as the evaluation metrics and datasets used. We also propose a taxonomy for interpretable semantic segmentation, and discuss potential challenges and future research directions.
Paper Structure (39 sections, 5 equations, 9 figures, 2 tables)

This paper contains 39 sections, 5 equations, 9 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Development of the Field of XAI in Computer Vision
  4. Specifics of Semantic Segmentation
  5. Limitations
  6. Taxonomy
  7. Scope: Local vs Global
  8. Method and its timing: Post-hoc vs Ad-hoc
  9. Range: Model-specific vs Model-agnostic
  10. XAI taxonomy for image segmentation
  11. XAI for Image Segmentation
  12. Methods
  13. Prototype-based methods
  14. Counterfactual explanations
  15. Perturbation-based methods
  16. ...and 24 more sections

Figures (9)

  • Figure 1: Publications with "explainable AI," "interpretable AI," and "AI regulation" as keywords. Publication data gathered from app.dimensions.ai
  • Figure 2: Explanation for single pixels: the selected pixels (top leftmost and centermost) are shown on the left, with their corresponding gradient-based explanations on the right.
  • Figure 3: Method-centered taxonomy
  • Figure 4: A framework for prototype-based methods.
  • Figure 5: A framework for counterfactual methods.
  • ...and 4 more figures