Introduction
Immunohistochemistry (IHC) has become a potent approach for detecting antigens in tissues. However, in current clinical pathology practice, interpreting IHC images is largely based on the personal experience of the expert pathologist, which is the major cause of different conclusions on the same IHC image among the pathologists, and such inter- and intra-observer variations could pose significant difficulties to clinical treatment-decision making [
1,
2]. Therefore, in the era of digital pathology, how to improve the accuracy and repeatability of quantitative analysis of IHC images and minimize the personal errors remains a formidable task [
2]. Previous study showed that image quality and color variations directly affect the result of image analysis made by pathologist or computer-aided image analysis [
3].
The conventional red–green–blue (RGB) images by common imaging technology are difficult to accurately distinguish labeled targets in the specific target region, due to inherent technical limitations including suboptimal image contrast and sharpness, strong light effect, overlapping chromogens, and varied staining intensities within or between chromogens, which directly affect the accuracy and repeatability in IHC image quantification and analysis [
4,
5]. A solution to this problem has to be found so as to improve the current IHC image reading and reporting system. However, an effective method to overcome this limitation is to use a multispectral imaging (MSI) system to acquire images and perform linear unmixing to separate and quantify each marker without interference from the others, irrespective of image overlap spatially [
4‐
6].
Recent developments in MSI systems such as the Nuance™ MSI system have greatly facilitated the imaging, visualization, and quantitative analysis of single-color or multicolor tissue samples, both in bright-field images and fluorescent images [
7,
8]. The advantages of MSI over traditional RGB imaging are to transform black–white or color RGB images with low contrast into spectral images with high color dimension, and obtain high-resolution images with precise spectral and spatial information [
9,
10]. Such spectral images could facilitate the extraction of specific chromogen information from the bright-field image according to spectroscopy principle, and such technical advantage could optimally remove background auto-fluorescence from the sample and significantly increase the signal-to-noise ratio (SNR) of the image to enhance the image quality for subsequent analysis [
10,
11].
In routine clinical histopathology, however, it remains unknown whether the application of MS images in quantitative IHC analysis could obtain higher sensitivity and specificity, although some promising results of MSI in automated quantitative scoring of IHC have been reported [
7,
8,
12,
13]. In this study, we used invasive breast cancer (BC) IHC images as the analysis object, and HER2, CK5/6, and ER were selected for IHC staining to represent interesting biomarkers at the cellular membrane, cytoplasm, and nucleus, respectively. The objective of our study is to investigate if the automated quantification of MS images of the biomarkers could increase the precision and reliability of image analysis and provide more information compared with conventional RGB images, and to develop a standard digital imaging system for IHC diagnosis in clinical pathology.
Discussion
Quantitative pathology is the development trend in clinical diagnostic pathology, which requires two types of key technology to achieve this goal. First, the imaging technology for digital pathology with high definition, high resolution, and high contrast should be developed. Second, the acquisition and analysis technology with high precision, high throughput, and high sensitivity should also be developed. Focusing on common parameters in BC cell membrane, cytoplasm, and nucleus as the research object, this study attempted to validate a kind of novel and standard IHC digital imaging technique.
For imaging technology, compared with colorized digital camera or monochromatic camera, MS imaging can obtain the information on color spectrum distribution at each pixel of a color image, whereas conventional RGB imaging can only obtain information on specific color distribution of the three available channels (red, green, and blue) [
13,
25], as demonstrated by spectral curves obtained in our results. Previous studies showed that MS imaging currently uses liquid crystal tunable filter (LCTF) technology to allow relatively narrow wavelength bands (10∼20 nm) to transmit wavelengths of light in the visible and near-infrared ranges, and spectral data of different wavelengths was acquired. It establishes a three-dimensional “cube,” with each spectrum corresponds to each pixel point in the cube, and classifies and separates the spectrum efficiently and accurately, which can significantly ameliorate the imaging process and suit multiple target imaging of both bright-field and fluorescence images. After the spectral data is acquired, a high-quality color image is generated from the image stack (or cube) and displayed [
9,
10,
13]. The corresponding RGB and MS images of HER-2, CK5/6, and ER IHC staining in this work were acquired using RGB imaging and MSI system. The imaging results showed that the overall quality of MS images is improved significantly, with richer color information, higher resolution, and stronger contrast, which will help quantitative image analysis. As suggested by Boucheron et al. [
26], MS images are better than the three standard bands of RGB images not only in color intensity, image resolution, and color contrast but also in additional useful information and image segmentation effect.
Previous applications of MSI in pathology mainly focused on analysis in routine hematoxylin-and-eosin images [
26‐
28], and the application of this technology in IHC analysis has been on steady increase. Traditionally, IHC interpretation may be standardized using current digital imaging technology, avoiding the use of three- or four-point subjective scoring methods. Several studies have shown that bright-field RGB scores were difficult to count exactly, as the tumor cells always overlap one another [
5,
6,
11,
29]. However, MSI can allow one to separate overlapping chromogens and unmix multiple chromogens in multicolor IHC [
11,
25,
30].
Accurate signal unmixing are prerequisites for valid quantitation of IHC-positive signal intensities. Quantitation typically requires some spatial manipulations such as segmentation of morphologic tissue and identification of subcellular compartments [
30]. In order to effectively analyze captured images, the software employed must be able to extract chromogenic intensity information from the images [
5,
30]. In our study, we observed the unmixed effect of different chromogens in RGB and MS images by CRi Nuance software package. It is simple to obtain clearer separated quantitative images and better segmentation accuracy from unmixing MS images of each marker than RGB images, no matter whether positive choromogens occurred in the membrane, the cytoplasm, or the nucleus. In order to visualize the co-localization better, these unmixed images can be recolored and layered together as composite images, rendering a simulated bright-field view. Therefore, the richer color information contained in MS images is more suitable for image quantitative analysis than conventional RGB images.
In BC patients, accurate assessment of hormone receptor (HR), HER2 status, and CK5/6 of the tumor is critical for defining the molecular subtypes and predicting response to systemic therapies [
21,
31,
32]. We measured the total signal OD values in RGB and MS images of membrane (HER2), cytoplasm (CK5/6), and nucleus (ER) in invasive BC using CRi Nuance software package. The results showed that the total signal OD values were higher in MS images than the corresponding RGB images, and the differences achieved statistical significance (
P < 0.01, for all). This possibly due to MS images contain stronger color intensities and more precise segmented capacities. However, whether high numerical quantitative results can obtain high diagnostic coincidence rate remains to be seen. Our previous studies also showed that MSI could be applied to QD-based fluorescence imaging to help improve tumor classifications and predict tumor prognosis [
15].
In the validation study using HER2 as a model, we compared the MS images analysis outputs, RGB images analysis outputs, and FISH results. ROC curve analysis, a useful method to evaluate the performance of diagnostic systems, showed greater AUC, higher specificity, and sensitivity in MS images than RGB images. These results indicate that the MS images could facilitate better diagnostic performance than RGB images. Huang et al. [
13] demonstrated that MSI technology is reliable for objective and high-throughput biomarker quantitation and colocalization study using chromogenic multiplexed immunohistochemistry.
By measuring RGB images, we found no significant difference between HER2 expression and age, menopausal status, tumor size, lymph node status, histological grades, chemotherapy, recurrence (P > 0.05). However, by quantifying MS images, we first reported that quantitative results of HER2 expression had relation to the lymph node status and histological grades of invasive BC; the total signal OD values were higher in lymph-node-positive than in lymph-node-negative tumors and in high-grade tumors than in middle-grade tumors (P = 0.02 and 0.04). It seems that more information could be provided by quantifying MS images of the protein expression in relation to differentiation of invasive BC.
In addition, consistency analysis results showed that quantitative evaluation of MS images of membrane (HER2), cytoplasm (CK5/6), and nucleus (ER) reached a better inter-observer agreement and reproducibility than RGB images. Although some factors may influence inter-observer variability such as application of software and observers themselves, this indicates that quantitative estimation of MS images of specific biomarker may be sufficient for reliable and robust measurement of IHC in invasive breast carcinoma.
Our study has several limitations. First, although 86 BC specimens were enrolled in our study, it finally included only 45 HER2-positive, 35 CK5/6-positive, and 40 ER-positive cases, which could result in possible sample selection bias and would have been more justifiable with more subjects. Second, the objective evaluation including mean squared error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity (SSIM) [
33] of different image types was not performed in our results on the ground that we primarily focused on comparing their sensitivity, specificity, and application effect in automated quantitative analysis of IHC. Additional studies are needed to identify the advantages of spectral image by calculating accurately the objective evaluation index of the image. Third, the analysis software used in this study could not eliminate the interference caused by three biomarkers expressed in the stroma, although such expression in the stroma is very little, and we are comparing two kinds of digital images under the same conditions. Fourth, despite higher sensitivity and specificity and more accurate quantification in MS images than conventional RGB images, we did not assess the relationship between quantitative analysis of different images and prognosis of invasive BC. Further prospective clinical studies are required to validate the advantages of MS images.
In conclusion, our study demonstrates the potential of MSI in quantitative analysis of IHC and highlights the inherent advantages of more accurate spectra quantification of MS images over conventional RGB images. Based on these advantages, an improved technical platform for IHC digital imaging could be developed for computer-aided quantitative digital pathology.