Mar. 6, 2008: In-Vivo Experiments

Mar. 6, 2008: In-Vivo Experiments

The control system, in its entirety, has been tested on in-vivo canine prostate tissue. Pre-operative imaging data of a canine prostate and the neighboring anatomy was taken and used to create a FEM mesh Figure 1. The pre-operative imaging data was taken at a resolution of 256x256x36 voxels with a field of view of 240mm x 240mm and an out of plane spacing of 1.5mm. The FEM mesh consisted of 23,303 dof and 20,064 hexahedral elements. A pipeline of commercial software was used to generate the mesh. AMIRA, www.amiravis.com, was used to perform a semi-automatic segmentation of the prostate and create a low resolution tetrahedral mesh of the prostate. CUBIT, cubit.sandia.gov, was used to convert the tetrathedral mesh into a hexahedral mesh and apply boundary conditions.

Figure 1: Pre-operative and intra-operative imaging data of canine prostate and the neighboring anatomy. The canine is laying on his back with legs upward. The pre-operative data was used to create a 3-D FEM mesh of the prostate consisting of 23,303 dof and 20,064 hexahedral elements. A cropped section of the FEM mesh is shown. The field of view shown in the imaging data is 240mm x 240mm. The resolution of the pre-operative and intra-operative data is 256x256x36 and 256x256x12, respectively. The relative difference in the location of the prostate is due to a bowel movement of the canine. The intra-operative image shown was used to locate the interstial laser fiber within the DICOM coordinate system.

Intra-operative imaging data is shown in Figure 1. The intra-operative imaging data was taken at a resolution of 256x256x12 voxels with a field of view of 240mm x 240mm and an out of plane spacing of 4mm. Comparison of the pre-operative image and the intra-operative image reveals that the prostate has moved a significant amount. The movement was uncontrollable, and due to a bowel movement that occurred sometime between the acquisition of the two data sets. The movement poses a significant registration problem. Current capabilities permit rigid registration only. The rigid registration is based on detecting the outer surface of the canine, not the prostate within. The rigid registration code found the out-of-plane position but failed in-plane due to the prostate movement. Consequently, under the given circumstances, the mesh was registered manually using AVS for interactive visualization. Furthermore, stringent time contraints on the experiment do not permit the use of the full resolution volume data sets for the rigid registration. The volume data sets must be subsampled to facilitate a reasonable exection time of the serial rigid registration code. However, for rigid registration, subsampling the imaging data by a factor of four in-plane does not loose much information on the outer boundary of the anatomy of the canine and significantly speeds up the registration from 25mins to 2mins with about 1mm difference in the final result. Another unexpected outcome of the registration process was that establishing connections between M.D. Anderson and TACC to send registration data was very time expensive. As a result, it is optimial to run the registration code locally at M.D. Anderson to minimize connections with TACC and only establish connections that are absolutely such as for the FEM computations.

The treatment day complications in the registration resulted in the use of the fail-safe structured mesh for the treatment day computations. The model calibration is dividing into a 180 second data acquisition phase followed by a 180 second computation phase. The canine prostate tissue was heated with a 3 Watt pulse for 90 seconds, 180 seconds worth of imaging is used capture the cooling of the in-vivo prostate tissue as well as the heating. The aquired imaging data is used for model calibration computations. Further treatment day complications resulted in poor quality thermal imaging data.

Figure 2: Depicted is the real-time visualization provided during the laser treatment. The anatomy, thermal images, FEM temperature prediction, and the power history are shown. Treatment day complications resulted in poor quality thermal imaging. Plenty of noise is seen in the thermal imaging but no disserable heating was detected during the calibration phase. The degree of noise prevalent is clearly seen in the cutlines of the thermal imaging. The power history shown was used in the treatment; it was computed for an uncalibrated model. A late time instance of the treatment is shown. The laser control of the uncalibrated model was manually overridden with the application of 9 Watts of power. The heating shown is due to the manual override.

The treatment imaging data shown in Figure 1 uses a space-time filter; in addition to the spatial median-deriche filtering pipeline, if the thermal data at a pixel changes by more than 11 C it is considered noise and filtered. The thermal data shows little heating as a result of the intial 3 Watt calibration pulse. Either, the amount of heating due to a 3 Watt pulse was within the noise range of the thermal imaging or the heating was quickly dissapated by blood perfusion. Consequently, the model calibration was unsuccessful as no disserable heating was detected in the thermal imaging data. The end result was the use of an uncalibrated model for the optimal control. The plan was to use the calibrated bioheat transfer model to create a 1.2cm lesion within the canine prostate. The optimal laser power as a function of time to heat the tissue region to 60 C for 300 seconds is computed. The power is controlled by the computers at TACC to create the lesion. However, as mentioned, the over filtering cascaded to the use of an uncalibrated model to control the heating of the tissue. The actual power history recorded by the visualase is shown in Figure 3. A time delay between the expected and actual power control is evident, as a result the initial 15 watt pulse of the optimal control did not occur. However, post-treatment computations show that the delay seen in power control of little consequence. The six second 15 Watts spike missing from the visualase log has an insignificant effect on the model predictions. Figure 3 also shows that copying the power control file directly to the mounted visualase directory interferes with visualase power log and instantaneously turns the laser off. The problem is a result of the file system architecture. Overwriting by copying the file instaneously deletes the file which instantaneously turns the power off. The fix is trivial, overwriting the file by moving does not delete the file and simply changes the file pointers. The sharp rise in power to 9 watts at the end of the time history of the visualase log file, Figure 3, is due to a manual override of the laser control due to the lack of heating seen. The predicted heating of the uncalibrated model was greater than seen in reality. Consequently, the model predicted power history was significanly less than needed to achieve the desired treatment plan.

Figure 3: A comparision between the expected and actual power history as a function of time is shown. The power history was extracted from the log files of the visualase for comparison with the expected power profile from the FEM code. The graph shows 5 second time delay between the expected laser control and the actual laser control. The is evident by the visualase’s log file failure to capture the 15 watt pulse at the beginning of the optimal control. The sharp rise in power at the end of the time history of the visualase log file is due to a manual override of the laser control due to the lack of heating seen.

The real time treatment data is posted below in 5 minutes interval.