Transformations are functions that map inputs to outputs by performing some operation, and consequently, image transformations map one set of pixels to another set of pixels, resulting in an output image that has been processed in some manner. ​​​Transformations are especially useful in medical data, as when training Artificial Intelligence on medical images, a lot of the usual rules of AI training break down. Some areas of the image may be useless or even counter-productive to take into account since they don't contribute anything towards detecting our goal. Meanwhile small differences in other areas of the image could result in entirely difference diagnoses, and thus we must make those minute changes prominent to the AI.

Transforming CT Scans

We'll start off by applying transformations to Computed Tomography scans (CT scans) of the brain. The following scans are of two types of patients: those diagnosed as normal, and those with a brain haemorrhage. Two of the images (CT scans of a haemorrhage and normal patient) are shown below.
Import["Haemorrhage-CT-Images/022.png"]
Out[]=
Import["Normal-CT-Images/136.png"]
Out[]=

ContourDetect

First we'll apply the ContourDetect function to the two images, with a delta of 0.8. The ContourDetect function returns a binary image where values of 0, i.e. the white pixels, correspond to the zeros and zero crossings of the image input. The delta parameter causes all values smaller than delta to be treated as zero.
ContourDetect[Import["Haemorrhage-CT-Images/022.png"],0.8]
Out[]=
ContourDetect[Import["Normal-CT-Images/136.png"],0.8]
Out[]=
As the images above show, applying ContourDetect to the CT scans results in a much simpler image which is much easier for AI to work with. It highlights haemorrhages because their white colour puts them above the delta while the grey matter is below the delta threshold.

ImageMesh

Another function that could be transform the scans, highlighting important features in the process, is ImageMesh. ImageMesh works by taking the nonzero pixels of an image as foreground.
ImageMesh[Import["Haemorrhage-CT-Images/022.png"]]
Out[]=
ImageMesh[Import["Normal-CT-Images/136.png"]]
Out[]=

GradientFilter + ColorNegate

Transforming PET Scans

MaxDetect

Conclusion