Case Study - Issues in Longitudinal Image Alignment

Here we present a Case study on using our in house skull strip and Slice Align software to provide the longitudinal measures of TCV (Total Cerebral Volume) and TCB (Total Cerebral Brain Matter). Subject 0-1183 underwent 4 brain MRI scans over 6 years. Traditionally each Brain scan is hand quantified by a skilled Neuroimager. We investigated a pseudo-automated quantification approach using our Skull Strip software coupled with Slice Alignment to use an existing (time1)image as a template to measure out the future volumetric quantities. The barriers to using a static template as a mask, is when the longitudinal images have a relative mismatch do to excessive movement. Alzheimer's or Parkinsonian tremors usually are causes of excessive slice to slice movement, as well as interleaving DSE image acquisition. Below is a study of the variance of traditional quantification and future methods employed by this lab.

Fig.1 Above is an MRI scan acquired coronally, and shown in Axial View to show the spatial movement in the patients left-right direction. The movement is charecteristic of an individual rotating on the occipital bone as a fulcrum, while laying in the MRI scanner(Alzheimer's Tremors). As seen on the gradient of increasing movement from Occipital to Frontal Areas. On the left is before Slice Align, on the Right is after. This tool enables the brain and surrounding tissues to align along Z-axis on a coronal image.

Fig.2 Here is a graph depicting the longitudinal volumetric quantities using three methods. Hand Trace is all manually traced TCV volume. Auto Trace is calculated with a Time1 mask linearally aligned to the TimeX, and Auto Trace with Slice align utilizes the slice Align algorithm to remove excess spatial movement between slices.

Preliminary results show that the use of Slice Align to remove the "dancing brain" phenomenom has reduced the variance of the calculated TCB tremendously as illustrated above. It is also important to show that even longitudinal brain stamping without Slice Align produced an impossible increase in TCB between serial series. Below is a summary of the variannce of each method. Further development of our automated volumetric analysis is needed as evidenced that even machines make mistakes.

For more details regarding the methods and other examples- please see this presentation

Up (Case Studies)