Explainable Artificial Intelligence for Breast Tumour Classification: Helpful or Harmful

Amy Rafferty, Rudolf Nenutil, Ajitha Rajan

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Explainable Artificial Intelligence (XAI) is the field of AI dedicated to promoting trust in machine learning models by helping us to understand how they make their decisions. For example, image explanations show us which pixels or segments were deemed most important by a model for a particular classification decision. This research focuses on image explanations generated by LIME, RISE and SHAP for a model which classifies breast mammograms as either benign or malignant. We assess these XAI techniques based on (1) the extent to which they agree with each other, as decided by One-Way ANOVA, Kendall’s Tau and RBO statistical tests, and (2) their agreement with the diagnostically important areas as identified by a radiologist on a small subset of mammograms. The main contribution of this research is the discovery that the 3 techniques consistently disagree both with each other and with the medical truth. We argue that using these off-shelf techniques in a medical context is not a feasible approach, and discuss possible causes of this problem, as well as some potential solutions
Original languageEnglish
Title of host publicationInterpretability of Machine Intelligence in Medical Image Computing: 5th International Workshop, iMIMIC 2022, Held in Conjunction with MICCAI 2022, Singapore, Singapore, September 22, 2022, Proceedings
EditorsMauricio Reyes, Pedro Henriques Abreu, Jaime Cardoso
PublisherSpringer, Cham
Pages104-123
Number of pages18
ISBN (Electronic)978-3-031-17976-1
ISBN (Print)978-3-031-17975-4
DOIs
Publication statusPublished - 7 Oct 2022
EventWorkshop on Interpretability of Machine Intelligence in Medical Image Computing at MICCAI 2022
- , Singapore
Duration: 22 Sep 202222 Sep 2022
https://imimic-workshop.com/

Publication series

NameLecture Notes in Computer Science
PublisherSpringer Cham
Volume13611
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Workshop

WorkshopWorkshop on Interpretability of Machine Intelligence in Medical Image Computing at MICCAI 2022
Abbreviated titleiMIMIC 2022
Country/TerritorySingapore
Period22/09/2222/09/22
Internet address

Keywords

  • Machine Learning
  • Breast Tumour Classification
  • Explainable AI
  • LIME
  • RISE
  • SHAP

Fingerprint

Dive into the research topics of 'Explainable Artificial Intelligence for Breast Tumour Classification: Helpful or Harmful'. Together they form a unique fingerprint.

Cite this