Preserving Dense Features for Ki67 Nuclei Detection

TL;DR

This paper presents UV-Net, a novel architecture designed to preserve dense nuclear features in pathology images, significantly improving Ki67 nuclei detection accuracy in breast cancer tissue analysis.

Contribution

Introduction of UV-Net, an architecture optimized for high-resolution feature preservation to enhance nuclei detection in dense and overlapping cell images.

Findings

UV-Net achieves an average F1-score of 0.83, outperforming other models.

UV-Net's accuracy exceeds other architectures by 9-42% on diverse datasets.

UV-Net demonstrates robust performance across multiple centers and unseen data.

Abstract

Nuclei detection is a key task in Ki67 proliferation index estimation in breast cancer images. Deep learning algorithms have shown strong potential in nuclei detection tasks. However, they face challenges when applied to pathology images with dense medium and overlapping nuclei since fine details are often diluted or completely lost by early maxpooling layers. This paper introduces an optimized UV-Net architecture, specifically developed to recover nuclear details with high-resolution through feature preservation for Ki67 proliferation index computation. UV-Net achieves an average F1-score of 0.83 on held-out test patch data, while other architectures obtain 0.74-0.79. On tissue microarrays (unseen) test data obtained from multiple centers, UV-Net's accuracy exceeds other architectures by a wide margin, including 9-42\% on Ontario Veterinary College, 7-35\% on Protein Atlas and 0.3-3\% on University Health Network.