gms | German Medical Science

Objective: Monitoring the volume of the Cerebrospinal Fluid (CSF) is desirable under therapy but very time consuming to assess. Common model based segmentation algorithms often have their limitations in gross anatomic changes as found in severe pediatric hydrocephalus. For this purpose, a more robust segmentation algorithm based on a hidden Markov random field model was implemented. We evaluated the CSF volume in an automatic fashion and estimated effects under therapy.

Methods: Retrospectively pre and postoperative true fast imaging with steady state precession (TrueFisp, 1mm isovoxel) were analyzed in 16 patients with pediatric hydrocephalus (male n=9, mean 3,6 ± 4,7 years, posthemorrhagic hydrocephalus n=3, hydrocephalus occlusus n=13). Patients were surgically treated (ventriculo-peritoneal shunt n=10, endoscopic third ventriculostomy (ETV) n=6). After preprocessing the 3D-datasets were skull stripped to estimate the inner skull surface, following a 3 class segmentation into different tissue types (brain matter, CSF) was performed. On the segmentation matrix the CSF-Volume was calculated.

Results: The method could be implemented in an automated fashion in all 16 patients. Mean CSF Volume at the beginning of the treatment was 384 ml ± 183 ml [163ml;824ml]. After surgical implantation of a ventriculo-peritoneal shunt CSF volume decreased significantly from 388 ml ± 169 ml [163ml;824ml] to 300 ml ± 114 ml [152ml;558ml]. Following an endoscopic third ventriculostomy CSF volume decreased significantly from 372 ml ± 205 ml [201ml;815ml] to 339 ml ± 201 ml [172ml;778ml].

Conclusion: A reliable segmentation could be performed with the implemented algorithm. The method was able to track changes in therapy.