gms | German Medical Science

Im 3D reconstruction is an important field in health care for medical use. In this work an application is presented which uses an extension of a novel method to obtain surface and volume data of human body for medical purposes. Basically the Photometric Stereo method is used. This technique uses four images of the same object taken under different light sources to obtain 3D information of the object. Instead of using normal light the light can be colored with the three colors of the RGB color model. Putting 2 cameras, facing each other, around the moving object and lighting the object with the colored light, a sequence of input images for each view direction can be obtained. After normalization and masking of each input frame, the volume data can easily be calculated. The obtained information can be used in certain medical fields such as rehabilitation, plastic surgery or orthopaedics.

In health care institutions 3D surface reconstruction for medical purposes improves continuously. Recently a new method was proposed to reconstruct the surface information of an object using the 3 Color Channel Dynamic Photometric Stereo Method [Ref. 1]. In 1980 Woodham presented the Photometric Stereo method to get the local surface orientation out of 2-dimensional pictures [Ref. 2]. The basic idea of this method is to take three or more images of an object with different illumination sources in known positions, but using the same camera. Then out of this information the surface of the object is reconstructed by calculating the normal vectors of every pixel, which leads to the 3D data by integrating. The disadvantage of this method is not being able to move the object while capturing the data, because the coordinates of every pixel in the obtained images of the object must be equal in each pixel. Using colored light instead of normal light sources the different colors can be filtered out of the image. Wentworth introduced in 1955 a standard in color television to describe a camera output pixel with the RGB system. Usually a pixels value can be separated into 3 color channels that range from 0 to 255. An image taken by a common color CCD camera separates the different channels. For the Photometric Stereo approach now 3 light sources having the 3 values of red, green and blue of the RGB color model can be used to light the object. Filtering each color channel out of the image, 3 grey level images can be obtained that represent the shadowed image created by each certain light source as it is used for the monochrome 3D reconstruction. Out of these 3 images the surface normal vectors can be calculated using the information of the light source directions and the CCD camera position and the assumed Lambertian Reflection Model for human skin. By putting two instead of one camera around the object we can acquire two image sequences of the patients movement. Before feeding the input frames to the algorithm there is a need to mask and normalize the frames. Using different colour models and difference images the patients body can be masked. This mask is used to restrict the reconstruction algorithm to significant regions in the image which decreases the computing time. Another important step is the normalization of the input sequences. As the reconstruction algorithm expects the object surface to be neutral coloured (white) but the patients skin is prevailing reddish, problems arise when feeding non-normalized images to this algorithm. To overcome these problems we use a calibration object to calculate the light distribution in the scene and a predefined skin sample to normalize the object colour. Regions with distortions within the masked area (nipples, belly button, spots of divergent skin colour) are masked again in order to being able to interpolate the reconstructed surface within the distorted regions. After these preprocessing steps the input data is ready for processing. Using the pre-processed input data from both cameras the reconstruction of the two halves of the patients body is possible. The reconstructed halves can be putted together by a merging and gap closing algorithm, yielding the volume of the whole patient body. Using a pixel-to-centimeter mapping the real volume of the patients body can be calculated. If the object or person is moving the complete movement process can be captured by using the Three Channel Photometric Stereo Method for every picture taken, resulting in a 4D patient volume model. This model is suitable for the measurement of lumbar spinal motion, lumbar lordosis and surface contour of back.

The presented method can be used to obtain the volume and surface of a real patient body. The surface information obtained with laser techniques are more precise compared to Three Color Channel Dynamic Photometric Stereo but the advantages are the lower price for the equipment and also the possibility of being able to move the object while capturing the data. In health care this techniques can be used in different fields such as rehabilitation, plastic surgery or orthopaedics. Using the Photometric Stereo technique for measurement of lumbar spinal motion, lumbar lordosis and surface contour of back was tested and lead to fact that the method is

suitable for medical use. Some psychiatric disease also causes movement disadvantages that may be analysed with the proposed frame.