Improved concordance of challenging human epidermal growth factor receptor 2 dual in-situ hybridisation cases with the use of a digital image analysis algorithm in breast cancer

Prognostic impact of HER2 biomarker levels in trastuzumab-treated early HER2-positive breast cancer

BACKGROUND

Overexpression of human epidermal growth factor receptor 2 (HER2) caused by HER2 gene amplification is a driver in breast cancer tumorigenesis. We aimed to investigate the prognostic significance of manual scoring and digital image analysis (DIA) algorithm assessment of HER2 copy numbers and HER2/CEP17 ratios, along with ERBB2 mRNA levels among early-stage HER2-positive breast cancer patients treated with trastuzumab.

METHODS

This retrospective study comprised 371 early HER2-positive breast cancer patients treated with adjuvant trastuzumab, with HER2 re-testing performed on whole tumor sections. Digitized tumor tissue slides were manually scored and assessed with uPath HER2 Dual ISH image analysis, breast algorithm. Targeted ERBB2 mRNA levels were assessed by the Xpert® Breast Cancer STRAT4 Assay. HER2 copy number and HER2/CEP17 ratio from in situ hybridization assessment, along with ERBB2 mRNA levels, were explored in relation to recurrence-free survival (RFS).

RESULTS

The analysis showed that patients with tumors with the highest and lowest manually counted HER2 copy number levels had worse RFS than those with intermediate levels (HR = 2.7, CI 1.4-5.3, p = 0.003 and HR = 2.1, CI 1.1-3.9, p = 0.03, respectively). A similar trend was observed for HER2/CEP17 ratio, and the DIA algorithm confirmed the results. Moreover, patients with tumors with the highest and the lowest values of ERBB2 mRNA had a significantly worse prognosis (HR = 2.7, CI 1.4-5.1, p = 0.003 and HR = 2.8, CI 1.4-5.5, p = 0.004, respectively) compared to those with intermediate levels.

CONCLUSIONS

Our findings suggest that the association between any of the three HER2 biomarkers and RFS was nonlinear. Patients with tumors with the highest levels of HER2 gene amplification or ERBB2 mRNA were associated with a worse prognosis than those with intermediate levels, which is of importance to investigate in future clinical trials studying HER2-targeted therapy.

Boosted Additive Angular Margin Loss for breast cancer diagnosis from histopathological images

Pathologists use biopsies and microscopic examination to accurately diagnose breast cancer. This process is time-consuming, labor-intensive, and costly. Convolutional neural networks (CNNs) offer an efficient and highly accurate approach to reduce analysis time and automate the diagnostic workflow in pathology. However, the softmax loss commonly used in existing CNNs leads to noticeable ambiguity in decision boundaries and lacks a clear constraint for minimizing within-class variance. In response to this problem, a solution in the form of softmax losses based on angular margin was developed. These losses were introduced in the context of face recognition, with the goal of integrating an angular margin into the softmax loss. This integration improves discrimination features during CNN training by effectively increasing the distance between different classes while reducing the variance within each class. Despite significant progress, these losses are limited to target classes only when margin penalties are applied, which may not lead to optimal effectiveness. In this paper, we introduce Boosted Additive Angular Margin Loss (BAM) to obtain highly discriminative features for breast cancer diagnosis from histopathological images. BAM not only penalizes the angle between deep features and their target class weights, but also considers angles between deep features and non-target class weights. We performed extensive experiments on the publicly available BreaKHis dataset. BAM achieved remarkable accuracies of 99.79%, 99.86%, 99.96%, and 97.65% for magnification levels of 40X, 100X, 200X, and 400X, respectively. These results show an improvement in accuracy of 0.13%, 0.34%, and 0.21% for 40X, 100X, and 200X magnifications, respectively, compared to the baseline methods. Additional experiments were performed on the BACH dataset for breast cancer classification and on the widely accepted LFW and YTF datasets for face recognition to evaluate the generalization ability of the proposed loss function. The results show that BAM outperforms state-of-the-art methods by increasing the decision space between classes and minimizing intra-class variance, resulting in improved discriminability.