Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Assessment of high- and low-risk histopathologic subtypes of basal cell carcinoma using artificial intelligence and dermatologist evaluation

Liang, Victor ; Shahrouki, Parasto ; Nejad, Daniel ; Backman, Eva ; Claeson, Magdalena ; Ingvar, Åsa LU orcid ; Krakowski, Isabelle ; Nielsen, Kari LU orcid ; Pakka, Jenna and Paoli, John , et al. (2026) In JAAD International 26. p.103-112
Abstract

Background: Basal cell carcinoma (BCC) is the most common skin cancer, and accurate histopathologic risk stratification is essential for treatment selection. Convolutional neural networks (CNNs) demonstrate promise in image analysis, yet few studies have evaluated classification of BCC subtypes. Objective: To assess the diagnostic performance of a CNN for binary classification of high-versus low-risk histopathologic subtypes of BCC and compare it with dermatologists. Methods: In this retrospective single-center study, 1580 dermoscopic images of histopathologically confirmed BCCs were used to train a preconditioned CNN. A test set (n = 249) was evaluated by 9 dermatologists. Performance was assessed using area under the receiver... (More)

Background: Basal cell carcinoma (BCC) is the most common skin cancer, and accurate histopathologic risk stratification is essential for treatment selection. Convolutional neural networks (CNNs) demonstrate promise in image analysis, yet few studies have evaluated classification of BCC subtypes. Objective: To assess the diagnostic performance of a CNN for binary classification of high-versus low-risk histopathologic subtypes of BCC and compare it with dermatologists. Methods: In this retrospective single-center study, 1580 dermoscopic images of histopathologically confirmed BCCs were used to train a preconditioned CNN. A test set (n = 249) was evaluated by 9 dermatologists. Performance was assessed using area under the receiver operating characteristic curve (AUC ROC). Logistic regression assessed associations between predefined tumor features and histopathologic subtype. Results: The CNN achieved an AUC of 0.75 (95% confidence interval [CI], 0.69-0.81) comparable to the dermatologists (AUC 0.79, 95% CI: 0.73-0.84, P = .18). Bumpy surface topography, clinical and dermoscopic ulceration and ill-defined border as well as dermoscopic focused vessels and white porcelain areas were associated with high-risk BCC, whereas dermoscopic unfocused vessels, erosions, and pigmentation were associated with low-risk BCC. Limitations: Retrospective single-center design and limited metadata for the CNN. Conclusion: CNNs appear useful for binary classification of BCC histopathologic subtypes.

(Less)
Please use this url to cite or link to this publication:
author
; ; ; ; ; ; ; ; and , et al. (More)
; ; ; ; ; ; ; ; ; ; ; and (Less)
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
artificial intelligence, basal cell carcinoma, convolutional neural network, machine learning
in
JAAD International
volume
26
pages
10 pages
publisher
Elsevier
external identifiers
  • pmid:42179893
  • scopus:105038280079
ISSN
2666-3287
DOI
10.1016/j.jdin.2026.03.014
language
English
LU publication?
yes
id
16440091-3b19-4a57-ac82-9680f54aa5d0
date added to LUP
2026-08-24 14:32:51
date last changed
2026-09-07 15:23:47
@article{16440091-3b19-4a57-ac82-9680f54aa5d0,
  abstract     = {{<p>Background: Basal cell carcinoma (BCC) is the most common skin cancer, and accurate histopathologic risk stratification is essential for treatment selection. Convolutional neural networks (CNNs) demonstrate promise in image analysis, yet few studies have evaluated classification of BCC subtypes. Objective: To assess the diagnostic performance of a CNN for binary classification of high-versus low-risk histopathologic subtypes of BCC and compare it with dermatologists. Methods: In this retrospective single-center study, 1580 dermoscopic images of histopathologically confirmed BCCs were used to train a preconditioned CNN. A test set (n = 249) was evaluated by 9 dermatologists. Performance was assessed using area under the receiver operating characteristic curve (AUC ROC). Logistic regression assessed associations between predefined tumor features and histopathologic subtype. Results: The CNN achieved an AUC of 0.75 (95% confidence interval [CI], 0.69-0.81) comparable to the dermatologists (AUC 0.79, 95% CI: 0.73-0.84, P = .18). Bumpy surface topography, clinical and dermoscopic ulceration and ill-defined border as well as dermoscopic focused vessels and white porcelain areas were associated with high-risk BCC, whereas dermoscopic unfocused vessels, erosions, and pigmentation were associated with low-risk BCC. Limitations: Retrospective single-center design and limited metadata for the CNN. Conclusion: CNNs appear useful for binary classification of BCC histopathologic subtypes.</p>}},
  author       = {{Liang, Victor and Shahrouki, Parasto and Nejad, Daniel and Backman, Eva and Claeson, Magdalena and Ingvar, Åsa and Krakowski, Isabelle and Nielsen, Kari and Pakka, Jenna and Paoli, John and Radros, Niki and Salmivuori, Mari and Polesie, Sam}},
  issn         = {{2666-3287}},
  keywords     = {{artificial intelligence; basal cell carcinoma; convolutional neural network; machine learning}},
  language     = {{eng}},
  pages        = {{103--112}},
  publisher    = {{Elsevier}},
  series       = {{JAAD International}},
  title        = {{Assessment of high- and low-risk histopathologic subtypes of basal cell carcinoma using artificial intelligence and dermatologist evaluation}},
  url          = {{http://dx.doi.org/10.1016/j.jdin.2026.03.014}},
  doi          = {{10.1016/j.jdin.2026.03.014}},
  volume       = {{26}},
  year         = {{2026}},
}