Brightfield Color Deconvolution with PCA ( IAColorDeconvolution )

Brightfield Color Deconvolution using principal component analysis (PCA) to convert RGB images to stain concentrations, stain concentrations to RGB images, normalize staining to standard and perform white balance adjustment. Select between these operations using the "Method" parameter. Select only one channel to run this function, the function automatically looks up all overlaid channels. It requires 3 channels with red, green and blue LUTs for both color to concentration or concentration to color operations.

* Converting H&E to concentrations:
Use "Method"="Color to Concentrations using PCA".
Use "Staining"="H&E" or "Two Stains...". The reference matrices do not affect conversion to concentrations, the stain vectors are automatically generated using PCA. When "Staining"="Two Stains - Calculate Referene" is used the Reference stain matrices will be saved as function parameter, so the "Method"="Concentrations to Color" and "Staining"="Two Stains - Calculate Reference" may be used for reverse operation, restoring the original brightfield image. The result replaces the original images, and the hematoxylin channel will be in the original red channel given in OD values, and the eosin in the green channel. The blue channel will return zeros.

* Converting Concentrations to bright field:
This is the reverse operation of converting brightfield to concentrations. Use "Method"="Concentrations to Color". Depending on the setting in the "Staining" parameter, a standard H&E vector, or the reference given in the "Reference OD matrix" or "Reference RGB matrix" will be used to create the colors. As input, 3 channels are needed, hematoxylin in the red channel given in OD values, and the eosin in the green channel, and the blue channel filled with zeros. The staining matrix can be saved from the original or another brightfield image using "Staining"="Two Stains - Calculate Reference" parameter, either with the normalization or color to concentration methods. Depending on the "White level" result images will be 8-bit or 16-bit per channel RGB images.

* Normalizing/standardizing H&E images:
Use "Method"="Normalize using PCA" and "Staining"="H&E" to recolor H&E stains to a built-in standard. Alternatively use the "Staining"="Two Stains - Calculate Reference" to calculate OD and RGB matrices using an image with the desired color, and then use "Two Stains - Use Reference" to normalize other images. The former operation writes the reference matrix parameters, the latter uses them.

"Reference OD matrix": maximal OD values for hematoxylin and eosin, written as ={{ODhematoxylin,ODeosin}}, or use a built-in kernel name.
"Reference RGB matrix": R, G, B coefficients to convert hematoxylin and eosin concentrations to colors, written as ={{R,G,B},{R,G,B}}, where the first set is for hematoxylin and the second is for eosin.

* Correcting white balance of bright field images:
Use "Method"="Auto white balance" and "Automatic white level"=Yes.
This method is applicable to RGB bright field images, where substantial amount of background, empty area is visible. The method will distort color if no enough background area is present. The image may have black or white empty (no data) areas at the edges, this will be detected as mask and optionally eliminated. Provide a new white level, that will be applied to each RGB channel in “White level”. “Automatic mask level” will decide if 0 or the bit-depth maximal value is used as no data points, and mask these areas during white balancing. Alternatively, manually apply mask level in “Mask level”. “Fill mask with white” will blend the masked area into the empty background area.

Based on: Marini N, Otalora S, Wodzinski M, Tomassini S, Dragoni AF, Marchand-Maillet S, Morales JPD, Duran-Lopez L, Vatrano S, Muller H, Atzori M. Data-driven color augmentation for H&E stained images in computational pathology. J Pathol Inform. 2023;14:100183.The "Stain vector file" is in R,R,G,G,B,B format with coefficients for H and E alternating for compatibility with the data in the above paper. Use data.astype(np.float32).tofile("HEref.bin") in Python to export stain vector database to binary format.