Abstract LB-279: Artificial intelligence aided diagnosis of breast cancer molecular classification based on immunohistochemical images

Abstract

Abstract Breast cancer has obvious heterogeneity. Molecular classification of breast cancer has been done through the study of gene expression profile. There were significant differences in gene expression, clinical characteristics, therapeutic response and prognosis among these molecular types, which were lumen A type, lumen B type, her-2 overexpression type and basal cell type, respectively. Compared with other nucleic acid level detection technologies, immunohistochemistry (IHC), as the most commonly used protein expression level detection method, has the advantages of low price, short cycle and more stable. IHC staining, including ER, PR, Ki-67, Her-2, CK5/6 and EGFR, can be used to classify breast cancer into different molecular subtypes and assist in the determination of treatment options. The application of artificial intelligence in pathological morphology is in the ascendant. Different algorithms can be used to evaluate both IHC staining and HE staining morphological evaluation of tumor tissues (including histological grading).In the molecular typing of breast cancer by IHC, the digital slide images obtained by scanning with software analysis can greatly reduce the subjectivity and instability of manual evaluation and improve the repeatability. Firstly, we scanned whole slide digital image of the HE staining slides and the six IHC slides of 200 surgical specimens of breast cancer with different molecular types (non-special types) from a medical center (Peking Union Medical College Hospital).Then compare the immunohistochemical results of each case obtained manually, adjust the algorithm, calibrate the parameters in the software, and finally get the fixed algorithm code and parameters. The algorithms used in study include image classification, nuclei segmentation, channel association and image registration. The algorithms were verified in the other surgical resection specimens of 400 cases of breast cancer with different molecular types (non-special types) from two other medical center, and was compared with the data obtained by manual reading. Finally, the results were encouraging, with the consistency of molecular classification reaching 100%. This study shows AI technology has great application prospect in pathological diagnosis of breast cancer. Citation Format: Jean J. Zhao, Minzi Ruan, Quancai Cui. Artificial intelligence aided diagnosis of breast cancer molecular classification based on immunohistochemical images [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr LB-279.