GPU-based system for performing 2D-3D deformable registration of a body organ using multiple 2D fluoroscopic views
Systems and methods for assisting a physician in a medical intervention comprises performing a 2D-3D deformable registration, and more particularly, performing a 2D-3D registration based on multiple live 2D fluoroscopic views, and implemented on a multi-core processing framework such as a Graphics Processing Unit.
1. A system for deformably registering pre-acquired 3D CT image data of a patient to real time 3D data of the patient comprising:
a first memory for receiving 3D CT image data of a body organ of the patient;
a second memory for receiving real-time image data of the patient from an imaging unit;
a multi-core processor framework that generates a candidate 3D deformation field based on one or more deformation parameters and the 3D CT image data; and
wherein the multi-core processor framework is further operable to compute a cost function and gradient vector to determine an optimal deformation field; wherein the cost function is based on the candidate 3D deformation field and the real time image data; and the gradient vector is based on the cost function.
2. The system of claim 1 , further comprising the imaging unit.
3. The system of claim 2 , wherein the imaging unit is one selected from the group consisting of a C-Arm, Biplanar fluoroscopy, 3D fluoroscopy, and Cone Beam CT.
4. The system of claim 1 wherein the multi-core processor framework comprises a CPU and GPU, and the optimizing step is performed on the CPU.
5. The system of claim 1 wherein the multi-core processor framework comprises a GPU, and the computation of the cost function and gradient vector are performed on the GPU.
6. The system of claim 5 wherein the multi-core processor framework is operable to implement a BFGS algorithm to optimize the candidate 3D deformation field.
7. The system of claim 4 comprising computing, in parallel, affected regions of the pre-acquired 3D data for a plurality of control points.
8. The system of claim 7 wherein the gradient vector further includes generating a first lookup table for pixel IDs in the affected regions for each control point.
9. A computer-implemented method for assisting a physician in a medical intervention on a patient comprising:
receiving pre-acquired 3D image data of a body organ of the patient prior to the medical intervention;
receiving real-time image data of the body organ; and
performing, on a multi-core processing framework, multiple parallel computations to determine a deformable registration on the pre-acquired 3D image data and the real-time data from the receiving step, and wherein the multi-core processing framework computes a 3D deformation field; and
wherein the performing step includes computing a cost function value and a gradient vector to determine the deformable registration on the pre-acquired 3D image data.
10. The method of claim 9 comprising inputting the cost function value and the gradient vector into an optimizing algorithm.
11. The method of claim 10 comprising computing, in parallel, affected regions of the pre-acquired 3D data for a plurality of control points.
12. The method of claim 11 wherein the gradient vector further includes generating a first lookup table for pixel IDs in the affected regions for each control point.
13. The method of claim 9 wherein the multiple parallel computations are carried out on a graphics processing unit (GPU).
14. The method of claim 10 wherein the optimizing algorithm is limited-memory BFGS based.
15. The method of claim 9 wherein the performing step is carried out in less than 30 seconds.
16. The method of claim 9 , wherein the receiving real time 3D data of the body organ is carried out with a unit for obtaining images.
17. The method of claim 16 , wherein the unit for obtaining imaging is one selected from the group consisting of a C-Arm, Biplanar fluoroscopy, 3D fluoroscopy, and Cone Beam CT.