AUTHOR=Xu Zilong , Yang Qiwei , Li Minghao , Gu Jiabing , Du Changping , Chen Yang , Li Baosheng TITLE=Predicting HER2 Status in Breast Cancer on Ultrasound Images Using Deep Learning Method JOURNAL=Frontiers in Oncology VOLUME=Volume 12 - 2022 YEAR=2022 URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2022.829041 DOI=10.3389/fonc.2022.829041 ISSN=2234-943X ABSTRACT=Purpose: Breast cancer molecular is critical for treatment guidelines such as the use of endocrine therapy. For breast cancer, human epidermal growth factor receptor 2 (HER2) is critical for treatment guidelines such as the use of targeted therapy. We developed a 3-block-DenseNet-based deep learning model to predict the expression status of HER2 in breast cancer in ultrasound images. Methods: We collected data from 144 breast cancer patients with pre-operative ultrasound images and clinical information from Shandong Province Tumor Hospital, retrospectively. An end-to-end 3-block-DenseNet deep learning classifier was built to predict the status of human epidermal growth factor receptor 2 separately in ultrasound images. The patients were randomly separated into a training (n=108) and a validation set (n=36). Results: Our proposed deep learning model achieved encouraging predictive performance in the training set (accuracy = 85.79\%, AUC = 0.87) and the validation set (accuracy = 81.56\%, AUC = 0.84). The effectiveness of our model outperformed the clinical model and the radiomics model significantly. The score of the proposed model showed significant differences in HER2 positive status and HER2 negative ones (p $<$ 0.001). Conclusions: These results demonstrate that ultrasound images are predictive of HER2 expression status through deep learning classifier. Our method provides a non-invasive, simple and feasible method for HER2 status prediction without manual delineation of region of interest (ROI). The performance of our deep learning model outperforms the traditional texture analysis based radiomics model significantly.