Finalized the internal volume compression: Caching mechanism during the voxel prediction stage, ROI compression optimization, final Huffman encoding of 32-bit data optimization.

Had initial discussion on femor segmentation with Dr. Rusinek, Gregory Chang and Alexandra Hotca.  Plan is to register given MRI volume, to a dictionary of femor shapes (binary ROI) constructed by an expert.  ISMRM  November submission is targeted

For 32-bit input Huffman encoding:  implementing a Frequency accumulator that does not require allocating a full Frequency table (too many entries for 32-bit) as it was in 8-,16-bit design.
Dynamically growing array of Leaves with binary search is implemented, this requires very little memory  This function is called in a heavily parallel workflow, so the memory efficiency is extremely important.

Working on internal compression of 3D\4D volumes.  Challenge here is that all PET timepoints have  different DICOM RescaleSlope tag.  So they have to be converted internally to pseudo real (floating point) format.   Predictive coding scheme is used. Resulting 32-bit residuals are compressed using Huffman.

Have to adapt and existing BYTE-oriented Huffman entropy coder to INT32 input