Image analysis of immunohistopathology from micro- to macro-scale: Application to ischemic brain tissue

Introduction

Immunochemical staining techniques are commonly used to assess neuronal, astrocytic and microglial alterations in stroke research. Conventional methods of analyzing immunohistology are based on image classification techniques applied to a specific anatomic location at high magnification. Such micro-scale (x10 or above) localized image analysis limits one for further correlative studies with other imaging modalities on whole brain sections. Direct feature extraction in macro-scale (x1 or x2.5) is not feasible. This report presents a semi-automated image analysis method that performs convolution-based image classification on x4 magnification images, extracts numerical data representing positive immunoreactivity, and creates a corresponding quantitative macro-scale image.

Materials and Experimental Description

Following 2 h middle cerebral artery occlusion (MCAo) by intraluminal suture model, rats were perfusion-fixed under halothane anesthesia following a 3-day survival period. Brain sections of 10 μm thickness were stained by EBA, lectin, GFAP or other methods to investigate cellular alterations, and viewed by a Nikon microscope equipped with a Sony 3CCD camera. The camera was interfaced to an image analysis system, MCID-M2. Each x4 magnification view was digitized and tiled by a motor stage controlled by the MCID-M2 software in a mosaic manner to form a TIFF format file for image analysis.

Methods

Image analysis for high magnification view is usually successful. For x4 magnification images, however, immunopositve pixels cannot be identified straightforwardly. We used several image-processing techniques to cope with variances in intensity distribution, as well as artifacts caused by light scattering or heterogeneity of antigen expression. Image classification was obtained by convolving the K-means clustering kernel over the tiled images. We then performed a morphological noise removal process (for single pixels or small area), called image closing. Following this process, another morphological filtering process, called image opening, was performed to filter unwanted target (large objects). In the last step of process, quantitative information was extracted and converted to a macro-scale image. Quantitative information includes the number of positively stained cells, number of positively stained vessels, or percentage area occupied by positively stained objects. This process is different from the “zoom out” function, which shows only averaged optical density of the corresponding pixel array in the micro-scale image, because it converts immunopositive objects within the corresponding window into numerical data; thus, the pixel intensity indeed represents a particular feature quantitatively.

Results

We applied the convolution-based K-means clustering method to the x4 images. The result was compared with that obtained from the x10 images by the conventional methods. For five areas covering the injured hemisphere, the conventional method estimated 1.37+/−0.22% immonopositve area in the corresponding hemisphere, and the designed method estimated 1.44+/−0.17%, both under anti-EBA staining; there was no significant difference between the two methods (p<0.05 ANOVA).

Conclusion

The result confirms that the convolution-based K-means clustering approach can be reliably applied to relatively low magnification images with considerable improvement in computational efficiency (compared to x10 images, 6.25-times processing time saved by using x4). The resultant quantitative macro-scale images can be correlated to other imaging modalities.