Spine project

See the presentation (in pdf format) made during the 15th Radiologist Research Symposium 

I work on this project with Dr Bullitt and Dr Ibanez. The Image-Guided Spine Surgery Project is a 5 years project. You can see a presentation of this project in the MIDAG (Medical Image Display and Analysis Group) webpage : spine surgery

Spinal operations are the most common surgical procedures. Indeed, arthritis is the most prevalent disease world-wide. The spinal operations require insertions of instruments into areas that surgeon cannot visualize directly. The standard of practice is to make decisions right. The cost of error is injury to nerve roots, spinal cord, or vascular structures.
MIDAG is working on a novel 3D-2D registration method that can map individual vertebra, as imaged by preoperative CT (Computer Tomography), to the patient by means of an intermediary, intraoperatively acquired fluoroscopic image. To reconstruct surgical instruments into 3D and to display them in their correct locations relative to the registered vertebra, the idea is to employ pairs of fluoroscopic images.
The general approach is to segment individual vertebrae fron the patient's CT scan and, through our registration process, to assign individual geometric transformation matrices to each vertebra.
Our approch take advantage of method to allow "articulate object" registration applicable to multiple vertebrae at one time,  instrument tracking, and intraopertive visualizations of the spine in its intraoperative configuration.

The patient’s position in preoperative CT scanner is different from the operative position, during the CT scan the spinal column is straight, but during the surgery the column is curved. A registration with the fluoroscopic image is needed during the surgery.

before after

Vertebra are extracted by first detecting intervertebral disk spaces. The space define a volume that contains each vertebra. Using a marching cube method we can extract a surface to represent each vertebra and display in its own coordinate system.

Using the Steven Aylward's program and code (VTree3D), we can extract the spinal canal like a tube. This extracted tube consists of a directed, central skeleton curve with a radius at each skeleton point. The program has been succefully apply for the spinal case but it require larger amount of computing time for data 512 by 512. So we need to subsampled the image to reduce it to 128 by 128.    

tube extraction

This figure figure the tube. This tube is the squeleton of the spinal cord. We can see that this tube is not straight. This curvatures have due to the space between two consecutive vertebrae, and the spinous process.

tree on view

Firstly, by smoothing the data with a gaussian filter and detecting the change of curvature of the tube, we can detect one plane for each intervetebral disk space.

disk without angle

We also implemented a method for computing the angle. So now we can extract the intervetebral disk spaces. Each purple plane represents an intervertebral disk space. We also put an arbitrary plane on top and on the bottom of the CT scan.

disk with angle

Using these planes, we can wrap each vertebra in a 3D bounding box. The lenght of the box is arbitrary. We use 80 mm here.




repre vert

one vertebra

We extract, using marching cube method, the vertebra inside the bounding box. The display is very fast, we can rotate the vertebra in real time, it is not the case for volume rendering methods.

vertebra march

The surgery take place in a room with two fluoroscopic units. A tool developped by Dr Ibanez is a reconstruction of the room with all the instruments and the controlled. This simulator was develloped for the Image-Guided Endovascular Neurosurgery Project.


We have exported the vertebrae extracted from the CT scan and reconstructed in the Angioroom. Here the 5 vertebrae in the table.    

Angioroom spine

These two screenshots of the fluoroscopic unit and the image taken by the simulator.

 angioroom fluo

spine fluo

Our current work is to build a simplify mechanical model of the spine.

This work was supported by grants R01CA67812 and P01CA47982 from NIH-NCI and an Intel equipment award.