PuSH - Publikationsserver des Helmholtz Zentrums München

Arcas, M.B.* ; Osuala, R. ; Lekadir, K.* ; Diaz, O.*

Mitigating annotation shift in cancer classification using single-image generative models.

In: (17th International Workshop on Breast Imaging, IWBI 2024, 9-12 June 2024, Chicago, US). 1000 20th St, Po Box 10, Bellingham, Wa 98227-0010 Usa: SPIE, 2024.:1317421 (Proc. SPIE ; 13174)
DOI
Artificial I ntelligence (AI) h as e merged a s a v aluable t ool f or a ssisting r adiologists i n b reast c ancer detection and diagnosis. However, the success of AI applications in this domain is restricted by the quantity and quality of available data, posing challenges due to limited and costly data annotation procedures that often lead to annotation shifts. This study simulates, analyses and mitigates annotation shifts in cancer classification in the breast mammography domain. First, a high-accuracy cancer risk prediction model is developed, which effectively d istinguishes b enign f rom m alignant l esions. N ext, m odel p erformance i s u sed t o q uantify t he impact of annotation shift. We uncover a substantial impact of annotation shift on multiclass classification performance particularly for malignant lesions. We thus propose a training data augmentation approach based on single-image generative models for the affected c lass, r equiring a s f ew a s f our i n-domain a nnotations t o c onsiderably mitigate annotation shift, while also addressing dataset imbalance. Lastly, we further increase performance by proposing and validating an ensemble architecture based on multiple models trained under different d ata augmentation regimes. Our study offers k ey i nsights i nto a nnotation s hift i n d eep l earning b reast c ancer c lassification and explores the potential of single-image generative models to overcome domain shift challenges. All code used for this study is made publicly available at https://github.com/MartaBuetas/EnhancingBreastCancerDiagnosis.
Altmetric
Weitere Metriken?
[➜Einloggen]
Zusatzinfos bearbeiten [➜Einloggen]
Publikationstyp Artikel: Konferenzbeitrag
Korrespondenzautor
Schlagwörter Dataset Shift ; Deep Learning ; Gans ; Image Synthesis ; Mammography ; Synthetic Data
ISSN (print) / ISBN 0277-786X
e-ISSN 1996-756X
Konferenztitel 17th International Workshop on Breast Imaging, IWBI 2024
Konferzenzdatum 9-12 June 2024
Konferenzort Chicago, US
Zeitschrift Proceedings of SPIE
Quellenangaben Band: 13174, Heft: , Seiten: , Artikelnummer: 1317421 Supplement: ,
Verlag SPIE
Verlagsort 1000 20th St, Po Box 10, Bellingham, Wa 98227-0010 Usa
Nichtpatentliteratur Publikationen
Begutachtungsstatus Peer reviewed
Institut(e) Institute for Machine Learning in Biomed Imaging (IML)
Förderungen Ministry of Science and Innovation of Spain
European Union