Deep Learning-Based H-Score Quantification of Immunohistochemistry-Stained Images

Abstract

Immunohistochemistry (IHC) is a well-established and commonly used staining method for clinical diagnosis and biomedical research. In most IHC images, the target protein is conjugated with a specific antibody and stained using diaminobenzidine (DAB), resulting in a brown coloration, whereas hematoxylin serves as a blue counterstain for cell nuclei. The protein expression level is quantified through the H-score, calculated from DAB staining intensity within the target cell region. Traditionally, this process requires evaluation by 2 expert pathologists, which is both time consuming and subjective. To enhance the efficiency and accuracy of this process, we have developed an automatic algorithm for quantifying the H-score of IHC images. To characterize protein expression in specific cell regions, a deep learning model for region recognition was trained based on hematoxylin staining only, achieving pixel accuracy for each class ranging from 0.92 to 0.99. Within the desired area, the algorithm categorizes DAB intensity of each pixel as negative, weak, moderate, or strong staining and calculates the final H-score based on the percentage of each intensity category. Overall, this algorithm takes an IHC image as input and directly outputs the H-score within a few seconds, significantly enhancing the speed of IHC image analysis. This automated tool provides H-score quantification with precision and consistency comparable to experienced pathologists but at a significantly reduced cost during IHC diagnostic workups. It holds significant potential to advance biomedical research reliant on IHC staining for protein expression quantification.

Keywords: H-score; deep learning; immunohistochemistry image; pathology image analysis; protein expression quantification.

Figure 1.

Flowchart of Automatic H-score Quantification for IHC Images.

Figure 2.

Examples of Region Recognition by the UNet-MobileNet Model Based on Hematoxylin Staining in IHC Images.

Figure 3.

Examples of Nuclei Masks Generated by Otsu’s Thresholding for Hematoxylin Staining in IHC Images, Along with Corresponding Tumor Cytoplasm Masks.

Figure 4.

Examples of Heat Maps Depicting the Distribution of DAB Intensity Categories Across Entire IHC Images, Determined by Selected Threshold Values.

Figure 5.. Comparison Between Manually and Automatically Calculated H-scores for the Development Set.

(A) Scatter plot demonstrating the correlation between the H-scores obtained through manual quantification by pathologists and those generated by our developed algorithm for the same IHC image. The black line represents the identity line, while the orange line is the regression line with a slope of 0.73 (95% CI: [0.70, 0.76]) and an intercept of −1.86 (95% CI: [−9.25, 5.52]). (We observed that the machine H-score tended to be lower than the pathologist H-score for the same IHC image, as explained in the discussion section.) (B) Scatter plot illustrating the correlation between the H-scores measured by two different pathologists for the same IHC image. The black line represents the identity line, while the orange line is the regression line with a slope of 0.62 (95% CI: [0.60, 0.65]) and an intercept of 98.75 (95% CI: [92.89, 104.60]). (C) Cumulative distribution function (CDF) of the coefficient of variation of H-scores for the IHC images assessed by pathologists or our algorithm from the same patient.

Figure 6.. Comparison Between Manually and Automatically Calculated H-scores for Two Independent Validation Sets.

(A) Scatter plot depicting the correlation between pathologist H-scores and machine H-scores of AKT3 for the 69 pancreatic cancer tissue samples in the first validation set. The black line represents the identity line, while the orange line is the regression line with a slope of 0.44 (95% CI: [0.36, 0.53]) and an intercept of 71.31 (95% CI: [51.48, 91.15]). (B) Scatter plot illustrating the correlation between pathologist H-scores and machine H-scores of HER2 for the 87 breast cancer tissue samples in the second validation set. The black line represents the identity line, while the orange line is the regression line with a slope of 0.62 (95% CI: [0.60, 0.65]) and an intercept of −0.39 (95% CI: [−4.27, 3.49]). (C) Scatter plot with error bars displaying the mean and standard deviation of the calibrated machine H-score for each tissue sample in the second validation set over 100 iterations. The black line represents the identity line.