MRI 3D image analysis and processing in orthopaedics
Fabrizio Ferrando (Esaote, Genova, Italy), Gianni Viano and Loris Vosilla (Softeco/Sismat, Genova, Italy)
ESAOTE is leading producer of low field MRI system which find large application in musculoskeletal examinations including, but not limited to, knee, shoulders, spine, wrist, hip and ankle.

Specific 3D image analysis techniques has been developed to provide radiologists and orthopaedist with advanced visualisation tools suitable to support diagnostic and surgery planning. These approaches are evolving toward objective parameter measurement to support Computed Aided Detection and Diagnosis in several fields including orthopaedics and rheumatology. To tackle this goal, ESAOTE is coordinating a joint public-private laboratory based in Naples, with the participation of academics, research institutes, medical centres and SMEs, for the identification, development and validation of dedicated CAD tools. This initiative exploits ESAOTE most innovative MRI products, G-SCAN, a tilting device allowing supine and weight bearing examinations. This feature enables the development of a new family of parameters obtained by comparing the differences between supine and weight bearing acquisitions, with the aim at defining new disease indicators and to support clinical follow up with quantitative data.

Within the several CAD tools under development, the “supine/weight bearing spine analysis” has a special interest for the 3D image analysis and visualization techniques involved. When changing from supine to weight bearing position, the spine of the patient under examination, which is a 3D physical object, has a movement comprising rotation and bending around the tree coordinate axes. The measurement of the variation from the two positions is the main objective of the CAD tool to provide indicator of the effect of weight on the spine and thus to provide objective parameters for pathological situation scoring and clinical evolution over time follow up. This measurement is a full 3D problem and require spine components 3D movement reconstruction, matching and registration to achieve accurate parameter computation.

The CAD tool developed starts from classical image analysis step: segmentation of body part of concern and 3D registration of images from different patient position The output is couple of registered 3D models of the spine which allow parameter measurement and comparison as well as direct visualisation of the difference to the user. Segmentation is based on customised level-set algorithms applied according to a set of heuristic rules which try to convey the a-priori knowledge about the anatomical districts considered. Relationships between body parts and their typical appearance on the image are used to impose segmentation constraints and to guide the process. A minimal support from the user is still requested, but a fully automated approach can be envisaged. 3D surface model is reconstructed from segmented image by using a “marching cube” approach. The parameters computed address the relationships among vertebrae and the shape and size of spinal cord. The comparison of parameter values in the supine and weight bearing position is supported as well.

3D model is used to provide realistic visualisation of the difference between the spine in the two attitudes. At this end 3D registration is carried out and a “merged” model is produced which allow to show the two models together and their relationships. A realistic virtual navigation environment is available as well to allow the user to explore the two models.