Things to watch in image analysis

Analysis of images captured on the fluorescence microscope allows for a great deal of information to be obtained from a set of pretty pictures! I'll probably come back to this theme quite a lot during these blogs as I come across people's work and update on the types of analysis people perform on their data.

Quantitative analysis of digital images essentially involves the characterization of pixels: intensity, size and proximity.

All digital images are made of pixels and these pixels reflect the brightness of the picture. Accordingly each pixel has a numerical value which reflects its brightness. These numerical values are on a scale referred to as a greyscale. The range of greys within the scale is reflected in the 'bit depth' of the image. 8 bit images have 256 greys (2 to the power 8) whilst 12-bit (2^12) and 16-bit (2^16) have more, and thus allow for more subtle variation to be reported. Larger bit depths are also larger file sizes. If you are collecting images to determine information such as shape and size bit-depth is not a factor.

Tip #1 - for image quantitation, especially intensity profiling, collect images in 12- or 16-bit depth.

Image saturation is the biggest killer of good quality image analysis. Saturation is essentially the predominance of either white or black pixels resulting in loss of detail in the image.

Tip #2 - ensure you use the whole dynamic range of the scale and avoid excessive levels of saturation - use range indicator look up tables to determine the presence of saturated pixels.

If you want to analyse images to count structures or determine the size or characteristics of a structure for example then you will need to threshold your data. This is often performed by moving a line on a histogram, with the line corresponding to the pixel intensities to be included in the analysis (right hand side of the line) and those to be excluded (to the left). This essentially removes background and isolates structures of interest. Within ImageJ you can then add these regions to a region of interest (ROI) manager for further analysis. The ROI amanger then allows your ROIs to be superimposed onto other images, counted and alalysed for whatever paramenter you are interested in (intensity, area, size, etc)

I've just briefly touched on some of the very basic aspects of image analysis. As I said earlier I will return to this as this subject plays a major role in modern imaging applications.