Detection of the Rotation Angle

The orientation procedure is anatomy-dependent. The procedure uses some anatomical features in order to determine the current orientation and necessary rotation angle. Three procedures are described for three anatomical structures: chest, pelvis, and hand/wrist.

In chest images the analysis is performed in three steps. First, the mediastinum is located and its orientation is found. This step excludes 90 rotations clockwise and counterclockwise in both groups of images (nonflipped and flipped). Then, a search for the subdiaphragm is performed. It eliminates upside-down images in both groups. Finally, images are tested against the y-axis flip.

Detection of anatomical structures (i.e., mediastinum, subdiaphragm, lungs) is based on average density measured within predefined windows scanned horizontally and vertically. The window size is determined by the width of the subdia-phragm assessed on the basis of clinical images in pediatric radiology. The average density measures yield average profiles reflecting the changes of the average gray-scale values in the horizontal (Fig. 8a) and vertical (Fig. 8b) direction.

In the horizontal direction, the mediastinum is marked by a high-value plateau placed between two low-value levels representing the lungs. The high-value plateau corresponds to a high-value average profile marked vertically in between two lower-value average profiles. One side of these average profiles increases, reflecting the subdiaphragm, which horizontally corresponds to a high-value plateau. Once the mediastinum and subdiaphragm are located, the images can be rotated to the upright position.

The final step detects the y-axis flip, recognized either by a detection of the local landmarks pointing the left or right image

FIGURE 8 Profile analysis in the detection of a current image orientation. (a) Horizontal profiles; (b) vertical Image can be at any orientation shown in Fig. 7.

profiles scanned over a CR chest image.

FIGURE 8 Profile analysis in the detection of a current image orientation. (a) Horizontal profiles; (b) vertical Image can be at any orientation shown in Fig. 7.

profiles scanned over a CR chest image.

side or by an analysis of the cardiac area. Landmarks (usually L for left or R for right image side) are placed within the radiation field, sometimes even within the patient contour. They also can be found (entirely or partially) in the area blocked by the collimator if only a sensitive image plate makes them visible. Their orientation (angle with respect to the image edge) also is random, and size may differ from one image to another. All this makes their detection and recognition more difficult. Standardization of the location and orientation of landmarks would make the task much easier.

Another way to detect the y-axis flip is an analysis of the cardiac shadow. Many approaches to the problem of lung segmentation and assessment of heart shadow size have been published already, and it will not be discussed in detail. They can be implemented at this stage of analysis. Also, a simple average profile analysis performed on the lower part of the lung yields two values that are referred to as a right cardiac index (RCI) and left cardiac index (LCI) and defined as (Fig. 9)

where a, b, c, and d are shown in Fig. 9. They reflect the size of the heart shadow in comparison to the overall lung size on both sides of the mediastinum. The LCI should be larger than the RCI. If this is not the case, the image is y-axis flipped.

FIGURE 9 Heart shadow analysis in detection of the y-axis flip.

Each anatomical structure requires its own function. For abdominal images, again horizontal and vertical average profiles are scanned. As for chest, the first stage locates the spine by searching for a high-value plateau. Then, perpendicular to the spine the average density and uniformity of upper and lower average profiles are tested. A denser area indicates the subdiaphragm. Location of spine and abdomen determines the rotation angle. No y-axis flip is considered.

For hand images, the analysis is performed on thresholded images (as discussed in Section 2.2). In order to find the correction angle, a pair of average profiles is scanned and shifted toward the image center until one average profile intersects the forearm, and the other at least three phalanges (Fig. 10). The forearm is detected if a high-value plateau located in the central part of the image and two neighbor low-value levels cover the entire image width (or height). Three phalanges are detected if three high-value plateaus are located within the average profile. The ranges of width of those plateaus are defined on the basis of anatomical width of forearms and fingers found in clinical pediatric hand images.

FIGURE 10 Pair of average profiles scanned over a CR hand/wrist image.

The search is completed when a pair of average profiles (scanned vertically or horizontally) meet the mentioned criteria.

0 0

Post a comment