Nuclei Classification in ER-IHC Stained Histopathology Images using Deep Learning Models
Wan Siti Halimatul Munirah Wan Ahmad, Md Jahid Hasan, Mohammad Faizal Ahmad Fauzi, Jenny T. H. Lee, See Yee Khor, Lai‐Meng Looi, Fazly Salleh Abas · TENCON 2022 - 2022 IEEE Region 10 Conference (TENCON) · 2022
Breast cancer treatment is highly dependent on the carcinoma stage, which was obtained by evaluating the pathological slides and the estrogen receptor status. The Allred score has been manually calculated by the pathologists to represent the percentage and intensity of tumor nuclei. The task can be automated by enabling digital pathology, by classifying the nuclei using learning-based method. We present here a comprehensive analysis of 32 pretrained deep learning models from DenseN et, EfficientN et, InceptionResN et, Inception, ResN et, MobileNet, NasNet, VGG and Xception. The aim of this exper-iment is to identify the best pre-trained model for classifying the negative, weak, moderate and strong nuclei taken from 44 whole slide images of estrogen receptor immunohistochemistry stained histopathology. The highest test accuracy is achieved by DenseNet169 with the measure of 94.91 %. This study will be a basis for the future development of more complex deep learning models with cascading or any combination of the tested models.