Prediction of HER2-positive breast cancer recurrence and metastasis risk from histopathological images and clinical information via multimodal deep learning

Jialiang Yang, Jie Ju, Lei Guo, Binbin Ji, Shufang Shi, Zixuan Yang, Songlin Gao, Xu Yuan, Geng Tian, Yuebin Liang, Peng Yuan

Research output: Contribution to journalArticlepeer-review

33 Scopus citations

Abstract

HER2-positive breast cancer is a highly heterogeneous tumor, and about 30% of patients still suffer from recurrence and metastasis after trastuzumab targeted therapy. Predicting individual prognosis is of great significance for the further development of precise therapy. With the continuous development of computer technology, more and more attention has been paid to computer-aided diagnosis and prognosis prediction based on Hematoxylin and Eosin (H&E) pathological images, which are available for all breast cancer patients undergone surgical treatment. In this study, we first enrolled 127 HER2-positive breast cancer patients with known recurrence and metastasis status from Cancer Hospital of the Chinese Academy of Medical Sciences. We then proposed a novel multimodal deep learning method integrating whole slide H&E images (WSIs) and clinical information to accurately assess the risk of relapse and metastasis in patients with HER2-positive breast cancer. Specifically, we obtained the whole H&E staining images from the surgical specimens of breast cancer patients, and these images were adjusted to size 512 × 512 pixels. The deep convolutional neural network (CNN) was applied to these images to retrieve image features, which were combined with the clinical data. Based on the combined features. After that, a novel multimodal model was constructed for predicting the prognosis of each patient. The model achieved an area under curve (AUC) of 0.76 in the two-fold cross-validation (CV). To further evaluate the performance of our model, we downloaded the data of all 123 HER2-positive breast cancer patients with available H&E image and known recurrence and metastasis status in The Cancer Genome Atlas (TCGA), which was severed as an independent testing data. Despite the huge differences in race and experimental strategies, our model achieved an AUC of 0.72 in the TCGA samples. As a conclusion, H&E images, in conjunction with clinical information and advanced deep learning models, could be used to evaluate the risk of relapse and metastasis in patients with HER2-positive breast cancer.

Original languageEnglish
Pages (from-to)333-342
Number of pages10
JournalComputational and Structural Biotechnology Journal
Volume20
DOIs
StatePublished - Jan 2022
Externally publishedYes

Keywords

  • Breast cancer
  • Convolutional neural network
  • H&E-stained histological images
  • HER2
  • Recurrence

Fingerprint

Dive into the research topics of 'Prediction of HER2-positive breast cancer recurrence and metastasis risk from histopathological images and clinical information via multimodal deep learning'. Together they form a unique fingerprint.

Cite this