THIS IS HOW DVIS WORKS.
DVIS algorithm does 3d reconstruction of backprojection efficiently. Back projection sampling and FFT is base of every 3d construction algorithm …the challenge is to do it fast enough so that it look like real time.
For example Qualcomm invented CDMA200, CDMAone ….but these are basically originated from well known CDMA concept. Qualcomm invented the most efficient way of using CDMA in cellular network …and they made billions of $...and still making. Even though it is made of basic CDMA concept …but it is still a patent ..and nobody can dare to use it without getting license from Qualcomm . Nobody thought of using this in cellular phone networking.
DVIS algorithm relies on the fact that a given object (internal body organ) can not move very fast 2/8 of seconds. Let say an organ 3d images consist of 64 voxel (smallest unit of 3d volume, like pixel in 2d picture) . At very first time DVIS scanner acquires all 64 samples. Now when scanner scans the organ second time, it is very likely that only few samples (let say 10-20) will change, unless patient is doing break dance inside the gantry.
When DVIS scans a organ , I guess there will initial latency of few seconds before organ images appears on the screen …but after that it will be continuous like real time , because SMART DVIS software only update the sample queue in incremental fashion ..i..e only sample which differs from the previous capture will be sent to 3d processor.
Each backprojection picture gives multiple pixel ..now multiple pixel (x,y coordinate) are processed to achieve 1 voxel (x,y,z coordinate).
Below is the very simple equation of 3d image
3d image (voxel 1,2,3,….N) = backprojection0 (pixel 1,2 3, N) + …backprojectionN (pixel 1,2 3, N) ;
3d reconstruction algorithm require lots multiplication, division and floating point operations. Any algorithm which reduces costly multiplication, division, and memory access can construct the image faster.
Let say a 3d images is consist of 16 samples. Let’s assume 3d processor needs to multiply these 16 numbers of get 1 voxel location. These samples are coming out of digitizer which converts analog output of flat panel detector into digital data.
Normal junk algorithm:
Int voxel = 1;
For(sample = 0 ; sample < 16; sample++)
{
Voxel = voxel * dig_out [sample];
}
Let assume each multiplication take 8 cycles.
Total cycle to get 1 voxel = 8 * 16 = 128;
Now let’s see how DVIS will work:
Assume DVIS pre-stored all 16 samples in a memory called base map after first scan. As I said this first scan might take some time …but I sure Deans must have dome something to even reduce this initial latency.
Superior SMART DVIS algorithm:
array base_map[8];
int last_voxel
int current_voxel;
int old_sample_mult = 1;
int new_sample_mult = 1;
for (sample = 0; sample < 8 ; sample ++)
{
if(base_map[sample] != dig_out[sample])
{
Old_sampe_mult = old_sample_mult * base_map[sample];
new_sampe_mult = new_sample_mult * dig_out[sample];
}
}
Current_voxel = last_voxel/old_samle_mult * new_sample_mult;
Let say only 2 sample were different from previous base map.
Total cycles = 2*8 + 8 (division) + 8 (last multiplication) = 32;
This itself makes DVIS 5 time faster (144/32) ..This ratio will increase as number of samples increases …actually they are millions of samples.
The above representation is highly simplified representation of DVIS algorithm …and this is based on my interpretation and understating of 3d reconstruction algorithm. Any valuable addition / comment in this discussion are always welcome.
Go IMGG ..will also make multi billion $ like Qualcomm …. ($5 Easy)
LongLoaded,
KingKong_Guru
