Atlas Based Anatomical Region Segmentation for Minimally Invasive Skull Base Surgery Objective Motion Analysis

Background: Surgical instrument motion objective analysis aims at improving surgical outcomes, objectively evaluating surgical skill, and improving trainee skill. Most existing objective surgical motion analysis schemes require structured surgeries or depend on recognition of motion patterns for certain categories of movement. Our previous studies found that anatomical regions played a significant role in statistical analysis of surgical instrument motion. Those studies showed that the statistical evaluation of whole unstructured surgical data lacks granularity for distinguishing differences in skill level. However, if we cluster the instrument motion data into anterior and posterior ethmoid and sphenoid regions, the statistical results were very different between expert and novice surgeons. These findings strongly suggest that more precise and finer anatomical region segmentation may lead to the finding of clearer surgical patterns for minimally invasive skull base and sinus surgeries. To achieve this goal, a precise and efficient semi-automatic atlas based skull base region segmentation algorithm was proposed and developed in the paper.

Methods: An efficient semi-automatic atlas-based skull base region segmentation method is proposed. An atlas was generated by manually segmenting the skull based area into 17 regions, including: Maxillary Sinus, Sphenoid Sinus, Sphenoid Bone, Foramen rotundum/V2 nerve, Vidian nerve/canal, Pterygopalatine fossa, Nasal airway, Frontal sinus, Skull Base 3 mm boundary, Nasolacrimal duct, Anterior ethmoid, Superior meatus, Posterior ethmoid, Inferior turbinate, Skull base, Orbit, and Middle turbinate. Other CT scans were segmented by aligning them with the atlas and registering the corresponding voxels. In the alignment, bony boundaries were first automatically generated by a new automatic bony intensity value detection based the region growing algorithm. Six features were then manually selected including: left eye center, right eye center, nasal frontal beak, anterior nasal spine, posterior edge hard palate, and dorsum sellae. The five landmarks formed a boundary of the skull base area and a landmark based affine registration algorithm was used to align the data with the atlas. Finally, the BSpline deformable registration was adopted to fine tune the registration.

Results: Eight CT scans from both female and male adults were segmented by this method using the original atlas. The segmentation results were manually inspected and the results showed that the segmentation error was 2 mm.

Conclusion: The proposed atlas based segmentation algorithm not only provides the desired precision, but also requires limited human interaction in the process. This method will contribute to automated segmentation of instrument motion data according to anatomical regions and more precise motion analysis will be achievable.

Fig: Example of region segmentation. Purple color indicate bony boundaries, other non-black areas correspond to segmented regions