Ki67 Proliferation Index Quantification using Silver Standard Masks

Seyed Hossein Mirjahanmardi, Melanie Dawe, Anthony Fyles, Wei Shi, Fei-Fei Liu, Dimitri Andtroutsos, Susan Done, April Khademi

Keywords: Deep Learning, Generalization, Computational Pathology, Nuclei Detection, Ki67

Abstract: Deep learning (DL) generalization for medical imaging remains an obstacle, fueled by the lack of enough available costly annotations. DL networks obtain high accuracy on datasets that are from the same training pool, but when applied to unseen datasets there is degradation in performance. This creates challenges for translation. This paper presents a self-annotation methodology to enhance generalization capabilities and reliability of deep learning architectures specifically developed for computational pathology images. We use a fully automated and unsupervised technique to generate silver standard (SS) masks for Ki67 nuclei on unseen datasets from the target hospital. A previously validated architecture for Ki67, UV-Net, is trained with gold standard (GS) images, a combination of SS and GS as well as SS alone to evaluate performance on the held-out test set. Preliminary results show higher accuracy of Diaminobenzidine (Ki67+) nuclei on the unseen data with the addition of the SS masks. SS masks provide an easy way to adapt models to new laboratories. While this technique is applied to the computation pathology datasets, other medical imaging domains will benefit from such approaches as well.

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.

Paper Type: both

Primary Subject Area: Learning with Noisy Labels and Limited Data

Secondary Subject Area: Application: Histopathology

Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.