Design of super-fast three-dimensional projection system for Positron Emission Tomography

Claim 2

An architecture for rapid 3-dimensional reconstruction of data collected from positron emission tomography (PET) scanners, said architecture being capable of selective back and forward projection, and said architecture comprising: a memory means for storing a 3-dimensional matrix to be imaged through back projection and for storing a complete set of two-dimensional views obtained through forward projection; a plurality of projection engines, each of said plurality of projection engines being used for storing a single two-dimensional view to be back projected and for storing a single two-dimensional plane of voxels from said three-dimensional matrix to be forward projected, each of said plurality of projection engines including a group of at least one Very Large Scale Integration (VLSI) chip and an engine memory, each of said VLSI chips being used for said back projection and said forward projection of said data, said engine memory being used for storage of data to be projected, each of said VLSI chips including an engine memory index calculator for computing a 3-dimensional coordinate transformation for projecting any point in said 3-dimensional volume to any point in said 2-dimensional projection plane and a bilinear interpolation calculator for computing a value P of a point of intersection of an LOR with said projection plane, said 3-dimensional coordinate transformation producing indices t i and u i of any position in said 3-dimensional volume into said 2-dimensional plane where t.sub.i =x.sub.i d.sub.ix +y.sub.i d.sub.ty +z.sub.i d.sub.tz ; and u.sub.i =x.sub.i d.sub.ux +y.sub.i d.sub.uy +z.sub.i d.sub.uz and where X i , y i , and z i represent the coordinates of said point in said 3-dimensional volume, and d tx , d ty , and d tz represent the cosines of angles defined between a t-axis of said 2-dimensional plane and x-, y-, and z-axes of said 3-dimensional volume, respectively, and d ux , d uy , and d uz represent the cosines of the angles between a u-axis of said 2-dimensional plane and said x-, y-, and z-axes of said 3-dimensional volume, respectively; and a dual controller for exchanging data between said memory means and a systolic array of multiple projection engines.

Claim 3

The architecture for rapid 3-dimensional reconstruction of data collected from PET scanners of claim 1 wherein said memory means is block random access memory (block RAM).

Claim 4

The architecture for rapid 3-dimensional reconstruction of data collected from PET scanners of claim 1 wherein said dual controller is a dual direct memory access (DMA) controller.

Claim 5

The architecture for rapid 3-dimensional reconstruction of data collected from PET scanners of claim 1 wherein said projection engines form a systolic array to accommodate parallel pipelining of data.

Claim 6

The architecture for rapid 3-dimensional reconstruction of data collected from PET scanners of claim 1 wherein said 3dimensional coordinate transformation equations are reduced to addition operations for unitary steps between voxels, said reduced equations being defined as: t j =t i +d tx ; and u j =u i +d ux for progression in said x-direction; t j =t i +d ty ; and u j =u i +d uy for progression in said y-direction; and t j =t i +d tz ; and u j =u i +d uz for progression in said z-direction, t i and u i being defined as the indices for a selected voxel v i , and t j and u j being defined as the indices for a voxel v j adjacent to voxel v i with voxel spacing equal to 1.

Claim 7

The architecture for rapid 3-dimensional reconstruction of data collected from PET scanners of claim 1 wherein said bilinear interpolation calculator determines said value P by bilinear interpolation from the four nearest Line of Response (LOR) count values and said 2-dimensional indices acquired from said engine RAM index calculator using the equation, and any algebraic derivations thereof: P=C+t f (D-C)+u f (A-C)+t f U f (B+C-A-D), t f and u f being said fractional components of t and u, respectively, and A, B, C, and D being the respective values of said four nearest LOR count values.

Claim 8

The architecture for rapid 3-dimensional reconstruction of data collected from PET scanners of claim 1 wherein each of said groups of VLSI chips is provided with at least a plurality of multipliers, adders, multiplexers, and register blocks for performing mathematical calculations on view data passed through said engine RAM index calculator and said bilinear interpolation calculator.

Claim 9

An architecture for rapid 3-dimensional reconstruction of data collected from PET scanners, said reconstruction of tomographic images being supported by an electronics board having an architecture capable of selective back and forward projection, and said board comprising: a plurality of I/O ports to allow for rapid and simultaneous data flow; a block RAM for the storage of a 3-dimensional matrix to be imaged through back projection and for the storage of a complete set of two-dimensional views obtained through forward projection; a plurality of projection engines, each of said projection engines being used for the storage of a single two-dimensional view to be back projected and for the storage of a single two-dimensional plane of voxels from said three-dimensional matrix to be forward projected, said projection engines forming a systolic array to accommodate parallel pipelining of data, each of said projection engines including a group of at least one VLSI chip and an engine RAM, each of said VLSI chips being used for the back projection and forward projection of said data, and said engine RAM for the storage of data to be projected, each of said groups of VLSI chips including an engine RAM index calculator for computing a 3-dimensional coordinate transformation for projecting any point in said 3-dimensional volume to any point in said 2-dimensional projection plane and a bilinear interpolation calculator for computing the value P of the point of intersection of said LOR with said projection plane, said 3-dimensional coordinate transformation using the simplified equations: t j =t i +d tx ; and u j =u i +d ux for progression in said x-direction; t j =t i +d ty ; and u j =u i +d uy for progression in said y-direction; and t j =t i +d tz ; and u j =u i +d uz for progression in said z-direction, t i and u i being defined as the indices for a selected voxel v i , t j and u j being defined as the indices for a voxel v j adjacent to voxel vi with voxel spacing equal to 1, and d tx , d ty , d tz , d ux , d uy , and d uz representing the cosine of the angles between the t axis of said 2-dimensional plane and the x, y, and z axes of said 3-dimensional volume, and the cosine of the angles between the u axis of said 2-dimensional plane and said x, y, and z axes of said 3-dimensional volume, respectively, said bilinear interpolation calculator determining said value P by bilinear interpolation from the four nearest Line of Response (LOR) count values and said 2-dimensional indices acquired from said engine RAM index calculator using the equation, and any algebraic derivations thereof: P=C+t f (D-C)+u f (A-C)+t f u f (B+C-A-D), t f and u f being said fractional components of t and u, respectively, and A, B, C, and D being the respective values of said four nearest LOR count values for back projection and voxel values for forward projections, and each of said groups of VLSI chips including at least a plurality of multipliers, adders, multiplexers, and register blocks for performing mathematical calculations on view and voxel data passed through said engine RAM index calculator and said bilinear interpolation calculator; and a dual DMA controller for exchanging data between said block RAM and a systolic array of multiple projection engines.

Claim 10

An architecture for rapid 3-dimensional reconstruction of data collected from PET scanners, said reconstruction of tomographic images being supported by an electronics board having an architecture capable of selective back and forward projection, and said board comprising: a plurality of I/O ports to allow for rapid and simultaneous data flow; a block RAM for the storage of a 3-dimensional matrix to be imaged through back projection and for the storage of a complete set of two-dimensional views obtained through forward projection; a plurality of projection engines, each of said projection engines being used for the storage of a single two-dimensional view to be back projected and for the storage of a single two-dimensional plane of voxels from said three-dimensional matrix to be forward projected, said projection engines forming a systolic array to accommodate parallel pipelining of data, each of said projection engines including a group of at least one VLSI chip and an engine RAM, each of said VLSI chips being used for the back projection and forward projection of said data, and said engine RAM for the storage of data to be projected, each of said groups of VLSI chips including an engine RAM index calculator for computing a 3-dimensional coordinate transformation for projecting any point in said 3-dimensional volume to any point in said 2-dimensional projection plane and a bilinear interpolation calculator for computing the value P of the point of intersection of said LOR with said projection plane, said 3-dimensional coordinate transformation using the simplified equations: t j =t i +d tx ; and u j =u i +d ux for progression in said x-direction; t j =t i +d ty ; and u j =u i +d uy for progression in said y-direction; and t j =t i +d tz ; and u j =u i +d uz for progression in said z-direction, t i and u i being defined as the indices for a selected voxel v i , t j and u j being defined as the indices for a voxel v j adjacent to voxel vi with voxel spacing equal to 1, and d tx , d ty , d tz , d ux , d uy , and d uz representing the cosine of the angles between the t axis of said 2-dimensional plane and the x, y, and z axes of said 3-dimensional volume, and the cosine of the angles between the u axis of said 2-dimensional plane and said x, y, and z axes of said 3-dimensional volume, respectively, said bilinear interpolation calculator determining said value P by bilinear interpolation from the four nearest Line of Response (LOR) count values and said 2-dimensional indices acquired from said engine RAM index calculator using the equation, and any algebraic derivations thereof: P=C+t f (D-C)+u f (A-C)+t f u f (B+C-A-D), t f and u f being said fractional components of t and u, respectively, and A, B, C, and D being the respective values of said four nearest LOR count values for back projection and voxel values for forward projections, and each of said groups of VLSI chips including at least a plurality of multipliers, adders, multiplexers, and register blocks for performing mathematical calculations on view and voxel data passed through said engine RAM index calculator and said bilinear interpolation calculator; and a dual DMA controller for exchanging data between said block RAM and a systolic array of multiple projection engines.

Claim 11

The architecture for rapid 3-dimensional reconstruction of data collected from PET scanners of claim 8 wherein each of said plurality of projection engines accesses at least two view memories such that a double buffering technique may be used to allow one set of RAM acquired from a set of said parallel LOR's to be accessed by said engine while a second set is loaded for a subsequent pass through an image, said controller swapping said access of said view memories at an end of each of said passes through said image data, said controller then loading a subsequent set of parameters into each engine.

Claim 12

The architecture for rapid 3-dimensional reconstruction of data collected from PET scanners of claim 1 wherein said engine memory is random access memory (RAM).