1 The program will accept and produce image stacks formatted as Bio-rad
confocal multi-image stacks. I am in the process of developing a Mac-based
image processing and analysis package which amongst other things converts
tiff stacks created by NIH image into BioRad confocal stacks. The package (called ViewPoint) is currently available only from me and costs 45 $US. A demo version is available. If interested I can be contacted at
(eric_s@biotek.mcb.uconn.edu).
2 The program is not friendly at all. This is largely to make it as portable as possible. It should run on any computer regardless of the machine architecture (ie it will check for files with both types of byte-ordering). It will currently read and write only Confocal image files saved as BYTE files. It will not run correctly with WORD files. I have run the program on several Macintosh computers, but it still takes about 6 minutes to render a single view of a "virtual volume" 256X126X105 on a Centris 650. I am currently running the program on an "Alpha" made by digital, where the same view will be calculated in 3.5 to 4 SECONDS. Some rendering "methods" will be faster than others. Rendering using a "shaded surface" method and simple expansions of the image planes will be fastest. Linear interpolation and transparent pixels generally produces the most useful results, but is computationally more costly.
3) Running the program. The program will simple die under several circumstances. In particular:
A) The program needs to hold the entire image volume in memmory (the "real" data, not the virtual volume). If it cannot allocate enough memmory to hold the data set it will exit abruptly.
B) The program will exit if the input image file is not an MRC-500/600 BioRad confocal image.
4) DIRECTIONS
The program first asks for the number of planes to interpolate between each image plane. This can be a real (floating pt) number such as 2.3 .
The program asks for the limit to z and the change in limit to z. This allows you to reconstruct a volume that has been "cut away" from the top. In addition, the change in z limit allows you to cut further into the volume for each view. Currently, the z limit applies to the original volume's z axis. Thus limiting a stack of 20 images by 4 means you will render only images 1 through 16.
The program then asks for the starting x,y, and angles. Sadly these must be entered in radians ergo 3.14 = 180 degrees. (I really should get around to fixing this). The program then asks for the step x,y, and z angles (again in radians) a step of 0.2 will produce a full rotation in about 31 views, while a step of 0.1 will take 62 views. For producing stereo pair images, rotate only in the y direction.
The program then asks for the number of extra pixels to calculate. The program (actually the programmer) is not smart enough to figure out where the limits of the volume are at each view. Because the volume is some sort of cube, as the volume is rotated the corners of the volume will extend beyond the dimensions of the original volume. The number of extra pixels entered will compensate by extending the volume that the program will search through to find objects. In other words, for a volume 100X100X100 spun on only 1 axis, the largest extent that will have to be searched is sqrt(100X100 + 100X100).
Next, the user is asked for the number of steps to calculate. This is the number of views around the object, and will also be the number of image planes in the final output image stack.
The program then asks for the method to be used in calculating the voluem between planes. Of these linear interpolation and plane expansion are the most commonly used and useful. Again linear interpolation is computationaly more costly.
The program asks for the method to be used in constructing the final image. Brightest pixel found is self explanetory. Transparency attempts to convey more information than simple brightest pixel. Essentially, if a bright pixel is found, overlaying pixels are allowed to make some contribution to the final image rather than simply being replaced. This method is greatly superior to an "average" value which I think makes everything look like mush, and is the mode I use most frequently.
The user will be asked for a transparency factor, which essentially determines how "transparent" objects will be. A transparency of 1 is the same as finding the brightest pixel.
Threshold mode will allow you to speicify that the first pixel value found within a specified range of values is what you are looking for. The program then asks for a weighting factor that will be used to calculate a "depth shaded" view of the volume. I usually use a value of 3 to 5 here.
Finaly, the program asks for the input image file name. If it finds the file, it will open it and attempt to load the file into memmory. If succesful it will ask for the name of a new output file. It will produce the message "starting image" for each image it calculates, so you can tell if the program has died or not.
NOTE: I have spent little time trying to make the program friendly. If the Mac OS really does start showing up on SGI's as promised by Apple, I will crank out a Mac interface pretty quickly. If you have comments, question or ideas regarding the program I would be happy to hear from you.
Regards, Eric.
APPENDIX: VIEW POINT I am currently developing an integrated software package for the analysis and enhancementof images in both Bio-Rad confocal format, standard tiff images, and multi-tiff image stacks produced by NIH image. The program will cost 45 $US. Among its present features are:
Project - using brightest pixel, darkest pixel, average pixel value, thresholded values using depth shading and transparent pixels.
Edge detection - using Laplacian, Robert Cross, Sobel and Kirsch operators. Resulting images are diplayed using original pixel values, the edge strength values, and as "binary" images. Edge detection operators can be thresholded to detect only well defined edges or all edges.
Linear feature detection - using semilinear and linear line detection algorithms. Resulting images can be diplayed as described for Edge detection.
Gray-scale surface (pseudo-3D) countour generation. Resulting images can be displayed using floating pixels, pixel expansion and linear interpolation.
Contrast enhancement - thresholding, base,scale,contrast stretch, inversion, chip and chop.
Filtering-Smooth, Sharpen, Gradient.
A graphics-style interface for color LUT generation, LUTs are directly compatable with NIH-Image.
Full-Screen viewing using bi-linear interpolated "zoom".
Inter-conversion between TIFF , NIH-Image "stack" and Bio-Rads Mrc500/600 multi-image files.
Object recognition and measurements including area, average pixel value and integrated pixel value
For more information, to obtain a demo version, or to become an authorized beta tester please contact me at eric_s@biotek.mcb.uconn.edu
