Three Dimensional Reconstruction with Linear CCD Cameras

Computerized tracking systems based on rectangular CCD cameras currently exist and a variety of commercial systems are available. One example application is the location of a surgical instrument on a CT or MRI image during brain surgery. The present study is based on developing a system of linear CCD sensors for tracking passive markers for the needs of neurosurgery. First the spatial resolution of such sensors is better, since they consist of up to 4096 pixels and second the amount of data to be processed is many times less than with rectangular CCDs. Our experiments showed good linear resolution of 0.1mm using a 1D sensor. The signal processing before passing the data for reconstruction also includes extraction of peaks from the signal and determining the peak positions along the sensor. An algorithm to supress the background is suggested, together with a routine to separate peaks located close to each other. The measurements taken from passive targets illuminated by infrared (IR) light and active IR emitting diodes demonstrate that the passive targets are well detected by the system. A three dimensional (3D) reconstruction algorithm, based on the Direct Linear Transformation (DLT), calculated (X,Y,Z) coordinates of the target points. The absolute maximum error observed in this case was 2mm. We conclude that an improvement of the accuracy is necessary at this final step before it is possible to use the system in combination with a CT or MRI image. Introduction There are several fields in medicine where three dimensional (3D) coordinate displacement is considered vitally important. These include: human movement analysis where patients with nervous, muscular or skeletal disorders are studied; radiation therapy and lithotripsy where the success of the treatment depends on the accurate positioning of the patient; and locating a surgical probe during brain operations. The need for exact positioning during these operations, after a lesion has been found tomographically, has always challenged neurosurgeons to place a sugical probe at the desired site. The interactive image-guided system is a device that allows the operating instrument to be tracked in real time and its tip to be visualized on a video screen, superimposed onto a tomographic image (CT or MRI) and pointing at the tumour site on the reference slice of the patient’s head (Maciunas, 1993). Such a system is feasible on the basis of the three dimensional (3D) reconstruction of space coordinates of the instrument’s tip. One method uses a number of projections from light markers attached to the end of the instrument, geometrically related to its tip and captured by sensors. These sensors could be one dimensional (1D) or two dimensional (2D) CCD cameras. In the case of 2D rectangular sensors, their number must be at least 2 to obtain the necessary data for the coordinate system of equations. If 1D linear sensors are used, the minimum number for a unique solution of the system is 3 (Macellari, 1983). We suggest the use of more than 3 linear sensors (between 4 and 8) and thereby creating some redundancy. The problem with multiple solutions will be solved using a least squares approach. Some of the steps to follow in analyzing and creating the described system, excluding the hardware design, can be defined as: (1) processing the raw CCD signal; (2) algorithm for peak position separation; (3) use of passive targets rather than active targets; and (4) algorithm for 3D reconstruction of the coordinates. Materials and Methods The test measurements comprise a number of experiments with active infrared light emitting diodes (IR LEDs) and passive (retro-reflective) markers illuminated by an IR source to check the intensity and the accuracy in positioning using either a cylindrical or a spherical lens. A rectangular CCD camera JVC TK1270, connected to a P360F Power Grabber (Dipix Technologies) and run on a Pentium 100MHz computer was used for this purpose. The captured images were stored in files with a picture size of 512x484 pixels and 8 bits pixel depth (28=256 gray scale levels).