Evident with the increasing cost and laborious work of diagnosis of cancer, using the gold standard Immunohistochemical (IHC) staining of tissues. Researchers from the University of California, Los Angeles, described a label-free virtual HER2 IHC staining based on a deep learning method.


Immunohistochemical staining of cancer tissues, such as the HER2 biomarker, is widely used to assess the degree of disease, analyze tissue, diagnose patients, explore pathogenesis, and guide treatment for cancer. Considering its capability to precisely identify the targeted biomarkers, IHC staining has been regarded as one of the gold standards for analyzing cancerous tissues and guiding treatment based on the diagnosis. Since its widely adopted for diagnosis, the overall time frame and infrastructure associated with the results of IHC staining are both expensive and extremely laborious.

Virtual HER2 staining of unlabeled tissue sections via deep learning.
Image Description: Virtual HER2 staining of unlabeled tissue sections via deep learning.
Image source: https://doi.org/10.34133/2022/9786242

The preparation of tissue samples and chemical processing requires professionals such as histotechnologist with increased timelines. Researchers from the University of California, Los Angeles, have described a deep learning method for label-free HER2 biomarker virtual IHC staining of breast tissue. The deep learning method uses a conditional generative adversarial network that is trained to swiftly transform the autofluorescence microscopic images of both unlabeled/label-free thin breast tissue sections into the bright-field or equivalent microscopic images. While the earlier research attempted and successfully created multiple types of histochemical stains, e.g., hematoxylin and eosin, and Jones silver stains. The earlier research focussed on developing the structural tissue stains while did not attempt to adjoin for virtual IHC staining.

Researchers displayed the method of virtual HER2 staining by training a dataset of 25 breast tissue sections using a deep neural network (DNN) sourced from 19 patients. A total of 20,910 image patches, each of 1024 pixels, were organized. Upon training the DNN model, virtual staining of the unlabeled sections of tissue was done using the autofluorescence microscopic images captured with dyes DAPI and FITC. Post-training, the team employed separate networking models to generate 12 virtual whole slide images (WSIs). Comparing both virtual and standard HER2 IHC staining of the breast tissue sections at different HER 2 protein scores. The results indicated that both the WSIs and the zoomed-in regions show high similarity between both virtual staining and standard IHC staining.

Results indicated that a well-trained virtual staining network such as DNN can reliably remodel the autofluorescence images of unlabeled breast tissue sections into the bright-field equivalent, virtual HER2 images that matched their standard IHC HER2 analogs. In order to blind evaluate and quantify virtual HER2, the team presented a framework with a blinded quantitative study. During the study, virtual whole slide images were jumbled across with the standard IHC HER2 whole slide images and presented to the board-certified pathologists who were grading scores HER2 score (i.e., 3+, 2+, 1+, or 0) without prior knowledge of the image source. Random shuffling and rotation of images were also applied to promote blindness in the study, and pathologists were asked to grade the image quality.

The summarized results for all the images from three pathologists resulted in the quality scores of virtual and standard IHC HER2 staining being in close proximity to each other and falling within their gold standard. As evaluated by the board-certified pathologists in order to determine whether the standard IHC HER2 images are both statistically and notably better than the virtual HER2 images in the staining quality. Out of two of the three pathologists reviewed, a statistically important improvement in the quality of the gold standard IHC staining compared to the virtual staining. While considering the virtually stained HER2 images did not mislead the diagnosis of breast cancer at the whole slide level.

Additionally, the research team followed up with a feature-based quantitative assessment of virtual HER2 staining, comparing the standard IHC staining. During the comparative analysis, a total of 8194 unique test image patches were blindly selected for the virtual training. Considering the different staining features of each different HER2 status, the blind testing images were divided into two subsets for the feature quantitative evaluation: subset one contained the images from HER2 0 and HER2 1+, = 4142, and the other subset contained the images from HER2 2+ and HER2 3+, = 4052.

Feature-based quantitative evaluation results for the virtual HER2 images compared against their standard IHC counterparts. The analysis demonstrated that the virtual HER2 staining feature metrics exhibited similar classification and closely matched average values compared to the standard IHC in terms of both the nucleus and the membrane stain features.

Final Thoughts

The study demonstrated a deep learning label-free virtual immunohistochemical staining method that clearly outperformed its predecessors. While the current gold standard IHC staining is efficient, conventional staining involves laborious chemical sample treatment steps demanding a histotechnologist’s make observations at regular intervals. The entirety of the process is deemed time-consuming, and hence, the presented trained DNN model produces the bright-field equivalent HER2 images computationally. It uses autofluorescence images obtained from label-free tissue sections. One of the advantages of the demonstrated method is its capability to generate highly uniform and repeatable results of staining while minimizing the staining variations that have been commonly observed in standard IHC staining. Hence, it provides the potential for better reproducibility for other diagnostic staining methods.

Article Source: Reference Paper

Learn More:

Top Bioinformatics Books

Learn more to get deeper insights into the field of bioinformatics.

Top Free Online Bioinformatics Courses ↗

Freely available courses to learn each and every aspect of bioinformatics.

Latest Bioinformatics Breakthroughs

Stay updated with the latest discoveries in the field of bioinformatics.

Website | + posts

Mahi Sharma is a consulting Content Writing Intern at the Centre of Bioinformatics Research and Technology (CBIRT). She is a postgraduate with a Master's in Immunology degree from Amity University, Noida. She has interned at CSIR-Institute of Microbial Technology, working on human and environmental microbiomes. She actively promotes research on microcosmos and science communication through her personal blog.


Please enter your comment!
Please enter your name here