HER2 (Human Epidermal growth factor Receptor) gene is a proto-oncogene, located on chromosome 17 q12–21.32, and is responsible for cellular proliferation, differentiation, and repair [1].

Through encoding a cell membrane tyrosine kinase receptor p185 HER2 oncoprotein (HER-2/neuorerbB-2); HER2 gene can stimulate cell proliferation through the RAS-MAPK downstream signaling pathway and inhibit cell death through the pathway of phosphatidylinositol 3’-kinase-AKT-mammalian target of rapamycin (mTOR) [2].

HER2 protein can be overexpressed in various types of cancer like breast, gastric, ovarian, colorectal and lung cancer, usually through increase HER2 gene copy numbers (amplification) [3].

Breast cancer (BC) is a heterogenous disease with five different molecular subtypes [4]. HER2 overexpression is identified in 18–30 % of primary BC patients, and is associated with rapid local recurrence, poor prognosis, shorter disease-free and overall survival (OS) in patients who don’t receive adjuvant therapy [5].

The accurate HER2 status assessment is essential to select BC patients who can benefit from the specific anti-HER2 targeted therapies such as Rastuzumab, Lapatinib, Pertuzumab, and other novel anti-HER2 targeted therapies [6]. All have been proved to improve HER2 +ve patients’ survival and outcome [7].

According to the American Society of Clinical Oncology/College of American Pathologists (ASCO/CAP) guidelines, the two approved methods for HER2 testing are immunohistochemistry (IHC) for detecting HER2 protein overexpression and in situ hybridization (ISH) assay for identifying the increased gene copy number [8]. Only patients who are HER2 score 3 + by IHC and/or those with amplified HER2 gene as confirmed by fluorescent in situ hybridization (FISH), benefit from anti-HER2 targeted therapies [6].

Several changes have been made on the definition of HER2 +ve status. Based on IHC; +ve HER2 status was initially defined by the Food and Drug Adminstration (FDA) as a uniform intense membranous staining in > 10 % of invasive BC cells [9]. In order to limit the false positive results, this cutoff was then changed by the ASCO/CAP group in 2007 to be > 30 % [10]. Later on, the revised and updated guidelines [ASCO/CAP 2013 and 2018] returned back the cutoff 10 % to maximize patients who are legible for targeted therapy and also to minimize false negative results [11], [12].

The definition of HER2 positive by ISH was determined initially by the FDA as having HER2/ chromosome enumeration probe 17 [CEP17] ratio ≥ 2.0, regardless of the average HER2 copy number) [9]. This was further expanded by the ASCO/CAP in 2007 and 2013 to include an average HER2 gene copy number ≥ 6 signals/nucleus regardless the HER2/CEP17 ratio [10], [11].

The updated guideline published by the ASCO/CAP group on 2018 classified patients into 5 groups. The working group addressed an algorithm with an addition workup for the less common ISH pattern (comprising about 4–15 % of cases), who have a mismatch between HER2/CEP 17ration and average HER2 gene copy number [12].

The development of an accurate and reliable machine learning (ML) model could improve the identification of HER2 positive BC patients, especially those with equivocal HER2 status, and facilitate the timely initiation of appropriate treatment. [13].

Therefore, in this study, we aimed to determine the clinicopathologic features which could predict HER2 positive BC patient. We also aimed to test the utility of AI-based ML in predicting HER2 status of BC.