Maintenance Notice

Due to necessary scheduled maintenance, the JMIR Publications website will be unavailable from Wednesday, July 01, 2020 at 8:00 PM to 10:00 PM EST. We apologize in advance for any inconvenience this may cause you.

Who will be affected?

Previously submitted to: JMIR Cancer (no longer under consideration since Jan 09, 2026)

Date Submitted: Jun 17, 2025
(closed for review but you can still tweet)

Prediction of KRAS, NRAS, BRAF, and HER2 Status in Colorectal Cancer Based on Histopathology Images Via Weakly Supervised Deep Learning

  • Xiang Zhang; 
  • Shuangshuang Wang; 
  • Qing Gu; 
  • Yuchen Fu; 
  • Hui Li; 
  • Jinwei Gan; 
  • Juan Du; 
  • Lele Chu; 
  • Xiuqing Li; 
  • Chenxi Wang; 
  • Li Li; 
  • Xuya Yuan; 
  • Xuya Yuan; 
  • Xuya Yuan; 
  • Xuya Yuan; 
  • Xuya Yuan; 
  • Xuya Yuan; 
  • Xuya Yuan; 
  • Yuan Li; 
  • Yi Zhang; 
  • Yifen Zhang; 
  • Yugen Chen

Background:

Research has shown that mutations in the KRAS, NRAS, and BRAF genes are linked to resistance to anti-EGFR therapies in colorectal cancer (CRC) patients. HER2-targeted therapies are increasingly being recommended for individuals with HER2 overexpression. The evaluation of KRAS, NRAS, BRAF, and HER2 statuses has become an important part of precise diagnosis for CRC. However, conventional molecular or protein testing can be time-consuming and expensive.

Objective:

This study aims to predict the status of KRAS, NRAS, BRAF, and HER2 through the analysis of whole-slide pathology features from CRC samples stained with Hematoxylin-Eosin (H&E) for KRAS, NRAS, and BRAF, and by utilizing Immunohistochemistry (IHC) for HER2.

Methods:

In this study, 435 CRC patients were enrolled from Jiangsu Province Hospital of Chinese Medicine. Using the clustering-constrained attention-based multiple-instance learning (CLAM) model, we constructed four models for predicting the statuses of KRAS, NRAS, BRAF, and HER2 based on whole-slide images (WSIs).

Results:

The dataset consisted of 435 WSIs from 435 primary CRC biopsy cases, with 298 included in each of the KRAS, NRAS, and BRAF models, and 157 cases included in the HER2 model. Our proposed four CLAM models demonstrated encouraging predictive performance, with all AUC values exceeding 0.88 (0.89-0.98). Our model-generated heatmaps showing KRAS, NRAS, BRAF mutation patterns and HER2 expression levels generally matched the regions identified by the pathologists.

Conclusions:

Our method provided new insights to predict gene mutations and protein expression using deep learning. These predictions could act as a prescreening tool, improving cost efficiency before the use of next-generation sequencing (NGS), amplification refractory mutation system-polymerase chain reaction (ARMS-PCR) and Immunohistochemistry (IHC). This approach ultimately enhanced the effectiveness of precision medicine and improves the consistency of quality in physicians’slide evaluations.

Clinicaltrial:


 Citation

Please cite as:

Zhang X, Wang S, Gu Q, Fu Y, Li H, Gan J, Du J, Chu L, Li X, Wang C, Li L, Yuan X, Yuan X, Yuan X, Yuan X, Yuan X, Yuan X, Yuan X, Li Y, Zhang Y, Zhang Y, Chen Y

Prediction of KRAS, NRAS, BRAF, and HER2 Status in Colorectal Cancer Based on Histopathology Images Via Weakly Supervised Deep Learning

DOI: 10.2196/79227

URL: https://preprints.jmir.org/preprint/79227

Download PDF


Request queued. Please wait while the file is being generated. It may take some time.