1 Introduction
Collagen fibers play an important role in sustaining structural integrity of healthy tissues. More crucially they play a vital role in pathological processes like inflammation and after tissue injury. For instance, chronic inflammation, irrespective of cause and organ affected, often results in fibrosis, i.e. an excess of connective tissue [
46]. This process is of particular importance in the heart as it is involved in numerous different cardiac diseases[
39]. In the diabetic or hypertensive heart disease for example, the myocardium of the left ventricle (LV) undergoes drastic structural remodeling. Activation of neurohumoral pathways, like TGF-β, endothelin-1, or Angiotensin II, stimulate the transition of fibroblasts into myofibroblasts and enhance the production of extracellular matrix (ECM) components in both ventricles in these hearts [
39,
48]. This not only affects myocardial structure, but also leads to stiffening of the ventricle and altered signaling to cardiomyocytes, known to contribute to both systolic and diastolic heart failure [
2,
39‐
41,
43]. Similarly, in pulmonary hypertension (PH), the right ventricle (RV) of the heart is subjected to pressure overload causing activation of fibroblasts and subsequent RV dysfunction [
15,
24]. Cardiac arrhythmias, like atrial fibrillation, are known to originate from a disturbed intercellular signaling due to fibrosis [
5,
12]. In many cardiac diseases, histological changes of the myocardium, including fibrosis, have been linked to cardiac function and even survival [
29,
40].
One of the most widely used methods to visualize fibrosis in histological tissue is by staining it with Picrosirius red [
21,
25,
32,
37,
45]. Picrosirius red, in contrast to more traditional stains like van Gieson and trichrome, has selectivity which makes it ideal and superior for both staining and quantification of collagen [
44,
45]. Moreover, known issues by traditional methods like staining variability and the tendency of stain to fade is absent when using Picrosirius red. Picrosirius red enhances the birefringence properties of collagen which makes the collagen fibers easy to detect using circularly polarised light [
13,
32,
45]. Quantification of fibrosis subsequently can be performed with stereology. This conventional manual planimetry method has proven its value but it is laborious, and the result depends in part on the skill of the observer.
Semi-quantitative and semi-automated methods of fibrosis quantification have been introduced [
3,
28,
45,
47]. However, detailed literature on methodological and technical aspects of fibrosis quantification is scarce. For a more precise analysis of the associations between extent of fibrosis on the one hand, and pathophysiology and clinical phenotype on the other, there is a clear need for better tools for quantifying fibrosis in tissue sections. To achieve this we investigated whether it would be feasible to quantify interstitial myocardial fibrosis using a semi-automated image analysis technique that would be reproducible, less sensitive to user interaction and reduces time needed for analysis. To this end, we employed the use of high-speed, automated whole slide imaging (WSI) system (also called digital slide system or virtual microscope) to convert ordinary glass slides into digital slides. Subsequent automated analysis of the captured images was carried out using a ImageJ macro-based algorithm that identifies, separates and quantifies Picrosirius red stain representing fibrosis.
4 Discussion
Introduced as prototypes only a few years ago, high resolution whole slide imaging (WSI) systems have become indispensable in the field of diagnostics and research alongside traditional microscopy. Despite their increasing potential, however, detailed literature comparing quantification technique using/between WSI and traditional microscope are scarce.
Our data shows that semi-automated quantification of fibrosis in histological samples obtained from WSI is feasible and strongly correlates with polarisation light microscopy and stereology. Although this method does not completely eliminates investigator’s influence, it reduces it to a minimum since determination of the optimal threshold was done only once by analyzing a test set of 63 images. In addition, our method is time-efficient, especially when large numbers of sections have to be analyzed, and it reduces investigator dependency to a minimum. The only investigator dependent aspect is the initial selection and acquiring of images in the regions of interest (RV, LV and septum), which applies to all methods studied.
Servais et al. [
38] in an earlier study reported the results of automated quantification of interstitial fibrosis in a trichrome stained sections of renal biopsies using color segmentation image analysis. However, detailed description of automated method of quantification and analysis were lacking making it hard to reproduce these results.
With respect to comparing standard methods to a new automated method, to our knowledge, only the study of Vasiljevic and co-workers, directly compared different histological methodologies used for fibrosis quantification [
42]. This study, on human endomyocardial biopsies, compares semi-quantitative scoring of fibrosis with computer assisted analysis and point-lesion counting using a grid. Vasiljevic et al. show correlations between different methods, however it must be noted these are to some degree subjective. In computer assisted image analysis, every section needs a threshold to identify fibrosis. In semi-quantitative scoring, the investigator itself has to determine the degree of fibrosis, again in every section. In our study, stereology correlated significantly with computer assisted quantification of fibrosis, which is in accordance with Vasiljevic et al. The major difference between their method and the method presented here is that our macro automatically discriminates between fibrosis and all non-fibrotic compartments, like lumen and cardiomyoctes. Moreover, artefacts can be excluded, although this requires manual annotations in the original images. Vasiljevic also did not apply polarisation microscopy, which according to literature, is still the method of choice.
Recently, Gaspard et al. published a detailed description of fibrosis quantification using colour subtractive image analysis in histological samples of myocardium [
14]. They described how thresholding of colours in digital images can be used to discriminate between fibrosis and surrounding tissue in a mouse model of isoproteronol induced myocardial hypertrophy. In this technique, the investigator has to set thresholds for every image that is analysed, which makes it sensitive for inter-observer variation and it is laborious. Compared to the study performed by Gaspard and co-workers, our method does not need thresholding in every image the investigator wants to analyze, since the threshold are set empirically once on a test set of images. However one may need a new set of test images for each scanner used or for each staining batch. In addition, their method was not validated by comparison with other methods used for histological fibrosis quantification.
Investigators using traditional stains like van Gieson and trichrome for detecting collagen are usually confronted with the typical problems associated with automated image analyses such as variations in staining intensity, hue, contrast and the stain’s tendency to fade. The fact underlying these problems is lack of precise selectivity for collagen fibers [
22,
34,
45] by these stains. In addition, qualitative studies have shown that collagen fiber structure changes with time as a function of age or as wound healing progresses [
8,
30,
44]. As expected a subsequent time associated shift in color spectrum (ranging from green to orange) would be found [
22]. In tissues where color change is expected to be present, separate hue (color) analyses may be necessary to study the structural details of collagen fibers.
To resolve these confounding issues, we chose Picrosirius red to stain collagen. Picrosirius red F3BA in saturated picric acid solution specifically and consistently stained thin collagen fibers, did not fade and was suitable for semi-automated quantification [
32,
34]. In addition we took further step in our staining technique by initially incubating frozen sections of myocardial tissues in xyline for 10 min as described earlier. This background eliminating process proves crucial in reducing staining variations to a minimum whilst generating a clear contrast between red stained collagen fibers in a bright yellow background of cardiomyocytes.
In our study, the ImageJ macro detected fibrosis in paraffin embedded tissue and on other tissues than frozen rat heart tissue as well. However, it must be noted that these tissues were examined with thresholds set for cardiac rat tissue. In principle the macro can be reused on any kind of histological staining in any kind of tissue, as long as structures in the macro are defined properly and thresholds are set accordingly.
Finally, it is worth noting that the ImageJ macro analysis detected more fibrosis than the polarization filter method (Figs.
4 and
5). This outcome is consistent with the study conducted by Diaz Encarnacion and colleagues [
11] in which collagen content was assessed and compared using 3 methods; Mason’s trichome, Picrosirius red with and without linearly polarized light. The highest collagen content was obtained in Picrosirius red stained sections without polarized light.
We scanned the sections with a Mirax scanner enabling us to digitally store histological sections for infinite period of time and for easy future references, allowing simultaneous viewing of whole slide histopathology and selection of regions for measurement. Essentially any image capturing device such as CCD camera could be employed for the acquisition of digital images of sections. Subsequent analytical measurements with our in house developed macro can be performed in an automated fashion by ImageJ. The macro can also be rewritten for commercially available software like ImagePro or Matrox Inspector.
We chose to investigate fibrosis in a rat model of PH, induced by monocrotaline, which is one of the most commonly used models to investigate RV pressure overload, and found a significant increase of collagen content in the RV of PH rats compared to controls. In our study however, fibrosis was not limited to the RV as it was also observed in the LV of these hearts. Lourenco et al. in an earlier study of monocrotaline induced rat model of PH also reported a similar observation [
24]. In addition, they also observed that RV fibrosis was significantly higher compared to LV. In our study however, we could not demonstrate such significant difference. Whereas Lourenco and co-workers sacrificed experimental animals after 6 weeks, ours were sacrificed after 3 weeks. Interestingly, in their previous work [
19] Lourenco and co-workers using the same experimental model reported an already established RV hypertrophy after 3 weeks. It is known in experimental models that right ventricular hypertrophy is accompanied by increased formation of myocardial fibrosis [
20,
26]. Moreover, as observed earlier, collagen fibers are expected to change as function of age or as wound healing progresses. It is also known that monocrotaline induced PH activates neurohumoral factors, which can also influence structural remodeling of the LV, in the absence of mechanical overload [
1,
23,
27]. In addition, vice versa, the RV also shows severe fibrosis in the setting of systemic hypertension [
4,
39,
47].