IHC-Net: A fully convolutional neural network for automated nuclear segmentation and ensemble classification for Allred scoring in breast pathology

Abstract

This paper presents a Deep Learning (DL)-based nuclear segmentation and ensemble classification scheme for quantitative evaluation of hormone status namely estrogen or progesterone on IHC specimen. This will mainly assist pathologists in automatic analysis and interpretation of breast cancer prognosis route. The proposed method consists of two major steps i.e., segmentation and classification. Since invasive breast cancer images are associated with numerous stained cells including artifacts like stromal and inflammatory particulars, it is crucial to develop a computerized method for segmenting them. A new segmentation method has been presented based on deep learning network for precisely segmenting out the stained nuclei region from breast tissue images. Morphological post-processing on segmented results shows the splitting of overlapped nuclei. Finally, to improve individual classifier’s results, the ensemble method is used, which integrates the decision of three machine learning (ML) models for final Allred cancer score. Statistical analysis reveals that all three classifiers perform adequately but proposed approach shows the best accuracy (98.24%), best correlation with the manual expert’s score (Pearson’s correlation coefficient = 0.908) and requires minimum computational time 44s/image (±2.33) compared to state-of-the-art methods. The proposed framework can be used as a reliable alternative to manual methods for automatic Allred scoring and in the prognostic assessment of breast cancer.