Interpreting and Correcting Medical Image Classification with PIP-Net

Meike Nauta*, Johannes H. Hegeman, Jeroen Geerdink, Jörg Schlötterer, Maurice van Keulen, Christin Seifert

*Corresponding author for this work

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

1 Citation (Scopus)
42 Downloads (Pure)

Abstract

Part-prototype models are explainable-by-design image classifiers, and a promising alternative to black box AI. This paper explores the applicability and potential of interpretable machine learning, in particular PIP-Net, for automated diagnosis support on real-world medical imaging data. PIP-Net learns human-understandable prototypical image parts and we evaluate its accuracy and interpretability for fracture detection and skin cancer diagnosis. We find that PIP-Net’s decision making process is in line with medical classification standards, while only provided with image-level class labels. Because of PIP-Net’s unsupervised pretraining of prototypes, data quality problems such as undesired text in an X-ray or labelling errors can be easily identified. Additionally, we are the first to show that humans can manually correct the reasoning of PIP-Net by directly disabling undesired prototypes. We conclude that part-prototype models are promising for medical applications due to their interpretability and potential for advanced model debugging.

Original languageEnglish
Title of host publicationArtificial Intelligence. ECAI 2023 International Workshops - XAI^3, TACTIFUL, XI-ML, SEDAMI, RAAIT, AI4S, HYDRA, AI4AI, 2023, Proceedings
EditorsSławomir Nowaczyk, Przemysław Biecek, Neo Christopher Chung, Mauro Vallati, Paweł Skruch, Joanna Jaworek-Korjakowska, Simon Parkinson, Alexandros Nikitas, Martin Atzmüller, Tomáš Kliegr, Ute Schmid, Szymon Bobek, Nada Lavrac, Marieke Peeters, Roland van Dierendonck, Saskia Robben, Eunika Mercier-Laurent, Gülgün Kayakutlu, Mieczyslaw Lech Owoc, Karl Mason, Abdul Wahid, Pierangela Bruno, Francesco Calimeri, Francesco Cauteruccio, Giorgio Terracina, Diedrich Wolter, Jochen L. Leidner, Michael Kohlhase, Vania Dimitrova
PublisherSpringer
Pages198-215
Number of pages18
ISBN (Print)9783031503955
DOIs
Publication statusPublished - 21 Jan 2024
Event26th European Conference on Artificial Intelligence, ECAI 2023 - Kraków, Poland
Duration: 30 Sept 20234 Oct 2023
Conference number: 26

Publication series

NameCommunications in Computer and Information Science
Volume1947
ISSN (Print)1865-0929
ISSN (Electronic)1865-0937

Conference

Conference26th European Conference on Artificial Intelligence, ECAI 2023
Abbreviated titleECAI
Country/TerritoryPoland
CityKraków
Period30/09/234/10/23

Keywords

  • Explainable AI
  • hybrid intelligence
  • interpretable machine learning
  • medical imaging
  • prototypes

Fingerprint

Dive into the research topics of 'Interpreting and Correcting Medical Image Classification with PIP-Net'. Together they form a unique fingerprint.

Cite this