AHSQ is a deep learning-based computational algorithm to quantify the H-score of immunohistochemistry (IHC) images. It starts with color deconvolution of the original IHC image to separate hematoxylin staining and DAB staining. Based only on hematoxylin staining, our trained UNet-MobileNet model is used for region recognition, and basic image processing methods are used for nuclei segmentation. By combining the mask for the target cell region and the mask for nuclei, the nuclear region and the cytoplasmic region of the target cells are identified respectively, and one of them is regarded as the target area for H-score quantification. The DAB staining of each pixel within the target area is classified as negative staining, weak staining, moderate staining or strong staining according to the predefined thresholds. The final H-score is calculated as a weighted average of the percentages of weak staining (weight = 1), moderate staining (weight = 2) and strong staining (weight = 3) within the target area. In sum, this algorithm can take an IHC image as input and directly output the H-score within a few seconds, which drastically speeds up the whole IHC image analysis procedure.
Prepare your IHC image file
Analyze the uploaded image with AHSQ online tool
Download reference images and summary file from the result page
Please cite our work if you find this AHSQ online tool helpful:
Zhuoyu Wen, Bret M. Evers, Danni Luo, Elena V. Daoud, Ruichen Rong, Shidan Wang, Kathryn A. O'Donnell, Yang Xie, Guanghua Xiao et al. "Deep Learning-based H-score Quantification of Immunohistochemistry-stained Images"