4D-CT medical compressible image processing using cellular neural network
Cáp Thanh Tùng, Phạm Thượng Cát · Journal of Computer Science and Cybernetics · 2012
With regards of 3D image, we use the concept of voxel(Volumetric cell) instead of pixel (picture cell).A single voxel consists of (x, y, z) for 3 space dimensions.Database of voxels can describe large structures and can be applied in many fields: architecture, video game, geology, astronomy, satellite image ... and, especially, in processing medical image.It is noted in the research of anatomy image from computed tomography of brain, lung, heart, thorax, hips... that those structures continuously and periodically change their position and size (due to respiration rhythm, pulse break, movement of muscles and articulations...) so the presence of time factor in parameter set of voxel is required.The processing of motion image concerns 4D-CT(4 Dimension Computed Tomography) scanning techniques and requires IR (Image Registration) techniques.There are many image registration approaches corresponding to various registration criteria (optical flow speed, calculation model, geometric feature ...).They are widely applied in analysis and calculation of medical images.Nevertheless, each approach still has issues to be addressed (both accuracy and calculation speed optimization).With medical image in general and with analysis of lung polyps diagnosis image, there is always considerable discrepancy.In general, the result will be rather good if intensity in image is unchanged , but this is impractical because lung is always in motion (due to respiration rhythm) and intensity on object (lung tissue) will be changed accordingly, which in turn leads to many errors in analysis.We call CT image of elastic tissues compressible image.Moreover, we investigate the calculation of optical flow speed of 4D-CT compressible image.This processing has a large number of image data and the calculation of optical flow speed is very complicated.So high processing speed is required and this is a great challenge for the PC.In this article, in order to improve calculation speed, we propose a processing with CNN for calculating 4D optical flow speed in real time mode. Tóm tȃ ´t.Vó .i a 'nh 3 chiê ' u, ta su .' du .ng khái niê .m voxel (Volumetric Cell) thay vì pixel (Picture Cell) cu 'a 2D.Mô .i phâ `n tu .' voxel do .n chú .a du .. ng các tham sô ´(x, y, z) cho 3 chiê `u không gian.Tâ .p ho .. p dũ .liê .u cu 'a các voxel có thê ' mô ta ' cho các hình thê ' ló .n, ú. ng du .ng trong nhiê `u lĩnh vu .. c: kiê ´n trúc, hoa .t hình (Video Game), di .a châ ´t, thiên vȃn, a 'nh vê .tinh,... và dȃ .c biê .t trong xu .' lý a 'nh y tê ´.Khi nghiên cú .u các a 'nh phâ ˜u thuâ .t có du .o . .c tù .các a 'nh chu .p cȃ ´t ló .p CT (Computed Tomography): não, phô ' i, tim, lô `ng ngu . .c, xu .o .ng châ .u... các khô ´i hình này còn liên tu .c thay dô ' i vi .trí và kích thu .ó. c theo chu kỳ (nhi .p tho .' , ma .ch dâ .p, su . .vâ .n dô .ng co ., khó .p...) nên câ `n tó .i yê ´u tô ´thò .i gian (t) trong bô .tham sô ´cu 'a voxel.Do dó viê .c xu .' lý a 'nh dô .ng liên quan tó .i kỹ thuâ .t scan 4D-CT (4 Dimention Computed Tomography).Quá trình xu .' lý a 'nh 4D-CT câ `n dê ´n kỹ thuâ .t xác nhâ .n a 'nh IR (Image Registration) mà ba 'n châ ´t là xác di .nh các diê ' m a 'nh tu .o .ng thích giũ .a hai a 'nh liê `n kê `trong quá trình chuyê ' n dô .ng. Có nhiê `u phu .o .ng pháp xác nhâ .n a 'nh khác nhau, ú. ng vó .i các tiêu chí nhâ .n da .ng khác nhau (theo tô ´c dô .cu 'a luô `ng a 'nh (Optical Flow), theo mô hình tính toán, theo dȃ .c diê ' m hình ho .c...).