Color correction of pathological images
Winner - Imaging InformaticsTokiya Abe ; Harvard Medical School;
Content:
The goal of the Multi-Spectral Imaging research is to develop decision support system for pathology. The color correction for H&E stained images by means of multispectral technique was proposed previously. However, it is difficult to introduce a camera system with large number of bands to general pathological institutions from the point of view of cost and operability. Therefore, this paper proposed a color correction method using a camera system with small number of bands.
Technology:
In this experiment, the multi-spectral imaging system which was utilized to capture the 16 bands images is composed of 2000 x 2000 pixels for the CCD camera, 16 band rotation filters, a conventional optical microscope Olympus BX-62 with an objective lens of 20 fold where the light source is a halogen lamp, and a PC based image capturing and displaying unit. The images with small number of band i.e. 3 bands images and 6 bands images were generated from 16 bands images.
Design:
In the color correction method, the transmittance spectra are estimated from multi-spectral images (MSI), and the amount of stain color pigment is calculated based on Beer-Lambert Law. It is possible to correct the color of tissue images of various staining conditions into an image of optimum staining condition by correcting the amount of coloring pigment calculated from MSI. Since the spectral characteristics of staining dye is affected by the staining condition, sufficient number of bands, probably greater than 3, is necessary to appropriately estimate the spectral transmittance of stained tissue components. Thus, this paper proposed a color correction method wherein the matrix used for the Wiener estimation is adaptively selected depending on the staining condition of the image.
Results:
Various H&E images of different staining conditions were corrected. The colorimetric error between the optimum stained tissue images and the corrected images reveals that the color of the corrected images closely matched to that of an optimum stained tissue image.
Conclusion:
The performance of the proposed method was investigated for H&E stained images with respect to different staining condition and number of camera bands. The results indicate that the proposed method performs better than the color correction method with conventional Wiener estimation.
