Post Hoc Interpretability of Deep Learning Models for Breast Cancer Histopathological Images with Variational Autoencoders

Muhammad Waqas; Tomas Maul; Iman Yi Liao; AMR AHMED

doi:10.1007/978-981-96-6948-6_7

Post Hoc Interpretability of Deep Learning Models for Breast Cancer Histopathological Images with Variational Autoencoders

Muhammad Waqas^*
, Tomas Maul
, Iman Yi Liao
, AMR AHMED

^*Corresponding author for this work

Computer Science

University of Nottingham Malaysia
University of Malakand
University of Nottingham

Research output: Chapter in Book/Report/Conference proceeding › Conference proceeding (ISBN) › peer-review

Abstract

Deep learning interpretability is very important, especially in medical imaging where clinical results can be greatly impacted by an understanding of model decisions. The application of variational autoencoders (VAEs) as post hoc interpretability tools for deep networks is investigated in this research. We show feature importance and examine model behavior through latent space perturbations by utilizing the latent space that VAEs have learned. The VAEs performance is evaluated and compared with numerous popular post-hoc interpretability techniques in this study for the Breast Cancer Histopathological Images. The comparison through RemOve And Retrain (ROAR) shows that the VAE based tool performs better in most cases and is the fastest among all the available post hoc interpretability methods.

Original language	English
Title of host publication	Neural Information Processing - 31st International Conference, ICONIP 2024, Proceedings
Editors	Mufti Mahmud, Maryam Doborjeh, Kevin Wong, Andrew Chi Sing Leung, Zohreh Doborjeh, M. Tanveer
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	91-104
Number of pages	14
ISBN (Print)	9789819669479
DOIs	https://doi.org/10.1007/978-981-96-6948-6_7
Publication status	Published - 22 Jun 2025
Event	31st International Conference on Neural Information Processing, ICONIP 2024 - Auckland, New Zealand Duration: 2 Dec 2024 → 6 Dec 2024

Publication series

Name	Communications in Computer and Information Science
Volume	2282 CCIS
ISSN (Print)	1865-0929
ISSN (Electronic)	1865-0937

Conference

Conference	31st International Conference on Neural Information Processing, ICONIP 2024
Country/Territory	New Zealand
City	Auckland
Period	2/12/24 → 6/12/24

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 3 Good Health and Well-being

Keywords

Breast cancer
Explainability
Interpretability
variational Autoencoders

Access to Document

10.1007/978-981-96-6948-6_7

Cite this

Waqas, M., Maul, T., Liao, I. Y., & AHMED, AMR. (2025). Post Hoc Interpretability of Deep Learning Models for Breast Cancer Histopathological Images with Variational Autoencoders. In M. Mahmud, M. Doborjeh, K. Wong, A. C. S. Leung, Z. Doborjeh, & M. Tanveer (Eds.), Neural Information Processing - 31st International Conference, ICONIP 2024, Proceedings (pp. 91-104). (Communications in Computer and Information Science; Vol. 2282 CCIS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-981-96-6948-6_7

Waqas, Muhammad ; Maul, Tomas ; Liao, Iman Yi et al. / Post Hoc Interpretability of Deep Learning Models for Breast Cancer Histopathological Images with Variational Autoencoders. Neural Information Processing - 31st International Conference, ICONIP 2024, Proceedings. editor / Mufti Mahmud ; Maryam Doborjeh ; Kevin Wong ; Andrew Chi Sing Leung ; Zohreh Doborjeh ; M. Tanveer. Springer Science and Business Media Deutschland GmbH, 2025. pp. 91-104 (Communications in Computer and Information Science).

@inproceedings{294733a785f84563b3520f56c1ce9256,

title = "Post Hoc Interpretability of Deep Learning Models for Breast Cancer Histopathological Images with Variational Autoencoders",

abstract = "Deep learning interpretability is very important, especially in medical imaging where clinical results can be greatly impacted by an understanding of model decisions. The application of variational autoencoders (VAEs) as post hoc interpretability tools for deep networks is investigated in this research. We show feature importance and examine model behavior through latent space perturbations by utilizing the latent space that VAEs have learned. The VAEs performance is evaluated and compared with numerous popular post-hoc interpretability techniques in this study for the Breast Cancer Histopathological Images. The comparison through RemOve And Retrain (ROAR) shows that the VAE based tool performs better in most cases and is the fastest among all the available post hoc interpretability methods.",

keywords = "Breast cancer, Explainability, Interpretability, variational Autoencoders",

author = "Muhammad Waqas and Tomas Maul and Liao, \{Iman Yi\} and AMR AHMED",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.; 31st International Conference on Neural Information Processing, ICONIP 2024 ; Conference date: 02-12-2024 Through 06-12-2024",

year = "2025",

month = jun,

day = "22",

doi = "10.1007/978-981-96-6948-6\_7",

language = "English",

isbn = "9789819669479",

series = "Communications in Computer and Information Science",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "91--104",

editor = "Mufti Mahmud and Maryam Doborjeh and Kevin Wong and Leung, \{Andrew Chi Sing\} and Zohreh Doborjeh and M. Tanveer",

booktitle = "Neural Information Processing - 31st International Conference, ICONIP 2024, Proceedings",

address = "Germany",

}

Waqas, M, Maul, T, Liao, IY & AHMED, AMR 2025, Post Hoc Interpretability of Deep Learning Models for Breast Cancer Histopathological Images with Variational Autoencoders. in M Mahmud, M Doborjeh, K Wong, ACS Leung, Z Doborjeh & M Tanveer (eds), Neural Information Processing - 31st International Conference, ICONIP 2024, Proceedings. Communications in Computer and Information Science, vol. 2282 CCIS, Springer Science and Business Media Deutschland GmbH, pp. 91-104, 31st International Conference on Neural Information Processing, ICONIP 2024, Auckland, New Zealand, 2/12/24. https://doi.org/10.1007/978-981-96-6948-6_7

Post Hoc Interpretability of Deep Learning Models for Breast Cancer Histopathological Images with Variational Autoencoders. / Waqas, Muhammad; Maul, Tomas; Liao, Iman Yi et al.
Neural Information Processing - 31st International Conference, ICONIP 2024, Proceedings. ed. / Mufti Mahmud; Maryam Doborjeh; Kevin Wong; Andrew Chi Sing Leung; Zohreh Doborjeh; M. Tanveer. Springer Science and Business Media Deutschland GmbH, 2025. p. 91-104 (Communications in Computer and Information Science; Vol. 2282 CCIS).

Research output: Chapter in Book/Report/Conference proceeding › Conference proceeding (ISBN) › peer-review

TY - GEN

T1 - Post Hoc Interpretability of Deep Learning Models for Breast Cancer Histopathological Images with Variational Autoencoders

AU - Waqas, Muhammad

AU - Maul, Tomas

AU - Liao, Iman Yi

AU - AHMED, AMR

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

PY - 2025/6/22

Y1 - 2025/6/22

N2 - Deep learning interpretability is very important, especially in medical imaging where clinical results can be greatly impacted by an understanding of model decisions. The application of variational autoencoders (VAEs) as post hoc interpretability tools for deep networks is investigated in this research. We show feature importance and examine model behavior through latent space perturbations by utilizing the latent space that VAEs have learned. The VAEs performance is evaluated and compared with numerous popular post-hoc interpretability techniques in this study for the Breast Cancer Histopathological Images. The comparison through RemOve And Retrain (ROAR) shows that the VAE based tool performs better in most cases and is the fastest among all the available post hoc interpretability methods.

AB - Deep learning interpretability is very important, especially in medical imaging where clinical results can be greatly impacted by an understanding of model decisions. The application of variational autoencoders (VAEs) as post hoc interpretability tools for deep networks is investigated in this research. We show feature importance and examine model behavior through latent space perturbations by utilizing the latent space that VAEs have learned. The VAEs performance is evaluated and compared with numerous popular post-hoc interpretability techniques in this study for the Breast Cancer Histopathological Images. The comparison through RemOve And Retrain (ROAR) shows that the VAE based tool performs better in most cases and is the fastest among all the available post hoc interpretability methods.

KW - Breast cancer

KW - Explainability

KW - Interpretability

KW - variational Autoencoders

UR - https://www.scopus.com/pages/publications/105009892690

UR - https://www.scopus.com/pages/publications/105009892690#tab=citedBy

U2 - 10.1007/978-981-96-6948-6_7

DO - 10.1007/978-981-96-6948-6_7

M3 - Conference proceeding (ISBN)

AN - SCOPUS:105009892690

SN - 9789819669479

T3 - Communications in Computer and Information Science

SP - 91

EP - 104

BT - Neural Information Processing - 31st International Conference, ICONIP 2024, Proceedings

A2 - Mahmud, Mufti

A2 - Doborjeh, Maryam

A2 - Wong, Kevin

A2 - Leung, Andrew Chi Sing

A2 - Doborjeh, Zohreh

A2 - Tanveer, M.

PB - Springer Science and Business Media Deutschland GmbH

T2 - 31st International Conference on Neural Information Processing, ICONIP 2024

Y2 - 2 December 2024 through 6 December 2024

ER -

Waqas M, Maul T, Liao IY, AHMED AMR. Post Hoc Interpretability of Deep Learning Models for Breast Cancer Histopathological Images with Variational Autoencoders. In Mahmud M, Doborjeh M, Wong K, Leung ACS, Doborjeh Z, Tanveer M, editors, Neural Information Processing - 31st International Conference, ICONIP 2024, Proceedings. Springer Science and Business Media Deutschland GmbH. 2025. p. 91-104. (Communications in Computer and Information Science). doi: 10.1007/978-981-96-6948-6_7