The document below reproduces poster #87 as presented at the Fifth International Conference of Functional Mapping of the Human Brain (HBM99) in Dusseldorf, Germany, June 22-26, 1999.
Reference: Neuroimage 9, Number 6, 1999, Part 2 of 2 Parts, p S87.
 

Semi-automated stripping of T1 MRI volumes: II. Blood vessel removal

Kelly Rehm 1,2, Kirt Schaper 2, Jeih-San Liow 1,2, Jon R. Anderson 2, David A. Rottenberg 1,2
1Department of Radiology, U. Minnesota, Minneapolis, USA, 2PET Imaging Center, VA Medical Center, Minneapolis, USA,

Abstract

Removal of blood vessels from stripped T1 MRI volumes is essential for creating high-quality 3D renderings and flat maps of the cerebrum and cerebellum. Automated and semi-automated algorithms for removal of scalp, skull and meninges from T1 MRI brain datasets often require manual editing to remove blood vessels due to lack of contrast between grey matter and vessels. If proton density (PD) and T1 volumes are acquired during the same scanning session, these volumes can be co-registered and bias-corrected PD intensity values used to identify and remove major blood vessels. 

Methods

T1 and PD MRI volumes are acquired during the same scanning session using a Siemens Vision 1.5T MRI scanner (T1: 3DFLASH protocol, PD: turbo spin dual echo protocol with PD and T2 weightings). There is minimal patient motion between the acquisition of the T1 and the PD volumes and little spatial distortion, however, PD volumes exhibit an intensity roll-off in the axial direction as well as significant in-plane non-uniformity. A strip mask is produced from a T1 MRI volume using a "consensus" stripping procedure (1) which combines intensity- and edge-based algorithms. This mask retains major blood vessels such as the sagittal sinus.

A rigid-body transformation matrix is calculated using AIR 3.0 (2) to co-register the T1 and PD volumes, and a dilated version of the initial T1 strip mask is used to produce a non-uniformity-corrected PD volume according to the method of Sled, et. al. (3). The uniformity-corrected PD volume then undergoes an inverse transform to bring it into alignment with the T1 volume and strip mask. Similar to the method used in T1 stripping, a threshold to distinguish low-intensity blood vessels and cerebro-spinal fluid (CSF) from grey matter is automatically selected based on the histogram of the uniformity-corrected PD volume.

Figure 1. Co-registered T1 and nonuniformity corrected PD volumes.

Methods (continued)

Figure 1 illustrates co-registered T1 and PD volumes for the same subject after application of the consensus mask that demonstrate the high PD contrast for vessels. Voxels below the threshold are subjected to a connected-component analysis; the largest surviving face-connected component consists of major vessels on the brain surface and a partial rim of CSF.

This vessel mask is deleted from the initial T1 mask, and the result is expanded slightly and smoothed by applying a 3D Gaussian filter (FWHM of 4mm) to the binary mask and thresholding the result. The latter step prevents minor misregistration of the T1 and PD volumes from improperly removing grey matter. The final result is the so-called "pruned" T1 strip mask (see Fig. 1, bottom row). In Figure 2 the major vessels pruned from each T1 mask are rendered in orange.

Figure 2. Vessels pruned from each test volume.

Results

Six repeat T1 and PD scans of a normal subject were used to examine the effect of vessel-pruning on strip masks; T1 volumes and strip masks for each volume were aligned to a common space for evaluation. The volume and extent (expressed as the percentage of the number of axial slices in the mask which were affected) of tissue removed are reported in Table 1. The effect of vessel removal on mask reproducibility was defined as the ratio of the number of voxels common to all masks divided by the number of voxels included within all six overlapping masks. Reproducibility increased with vessel pruning from 0.93 to 0.94.


 

Table 1.  Effect of automated vessel pruning.
Scan #
Vessel Volume (cc)
Slices pruned (%)
1
30.5
90
2
40.1
99
3
44.1
96
4
45.9
99
5
30.3
97
6
41.7
99
Average
38.8
96.7

Conclusions

Automatic removal of major blood vessels and superficial CSF from the strip mask of a T1 MRI volume is possible if a proton density volume is available and a good registration of the PD and T1 volumes can be achieved. The tissue volume removed automatically is small relative to the entire mask volume (about 6% of the consensus mask) but is distributed across 90 to 99% of all axial slices in the volumes studied -- a potentially significant reduction in effort for an operator performing manual pruning.


This work has been supported by National Institutes of Health grants NS33718 and MH57180.

References

  1. Rehm K, Shattuck D, Leahy R, Schaper KA, Rottenberg DA. "Semi-automated stripping of T1 MRI volumes: I. Consensus of intensity- and edge-based methods." Neuroimage 1999, 9(6) Part 2: S86.
  2. Woods RP, Grafton ST, et. al. J. Computer Assisted Tomography, 1998, 22(1):139-52.
  3. Sled JS, Zijdenbos AP, Evans AC. IEEE Transactions on Medical Imaging, 1998, 17(1):87-97.