Image analysis – Applications – Biomedical applications
Reexamination Certificate
1997-12-18
2001-01-16
Mehta, Bhavesh (Department: 2721)
Image analysis
Applications
Biomedical applications
C382S156000
Reexamination Certificate
active
06175643
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to Magnetic Resonance (MR) image displays, and more particularly, to an auto-windowing system for MR images capable of adapting to optimal display parameters, personal preferences and different viewing conditions while online.
2. Prior Art
Adaptive and automatic adjustment of the display window parameters for Magnetic Resonance (MR) images under different viewing conditions is a very difficult and challenging problem in medical image perception. There are several factors that make this problem difficult, namely, the function describing a human expert's adjustment of display window parameters for a wide variety of MR images is extremely complicated, the adjustment result is subjective and substantially depends on personal preference, the adjustment function varies with the viewing conditions, etc. The viewing conditions are generally the condition of the monitor and the exterior viewing environment, such as the illumination condition. It is almost impossible to account for all these issues in a single universal algorithm.
The display windowing process is primarily used in the mapping of 12-bit MR image intensity values into 8-bit gray levels for displaying MR images on a common 8-bit computer monitor. The display window consists of a width parameter and a center parameter. The windowing process maps the image intensity values linearly from [center-width/2, center+width/2] to the nearest integer in [0,255]. For the MR image intensity values below (center-width/2), they are mapped to 0. Similarly, the image intensity values greater than (center+width/2) are mapped to 255. Apparently, these two parameters can greatly influence the appearance of the image to be displayed. In other words, the brightness and the contrast of an image is determined by these two parameters. Inadequate adjustment of these parameters can lead to degradation of image quality, and in severe cases to loss of valuable diagnostic information of the images.
Most previous methods for the adjustment of display window parameters are either very restricted to certain types of MR images or perform very differently from the human adjustment. R. E. Wendt III, “Automatic adjustment of contrast and brightness of magnetic resonance images”,
Journal of Digital Imaging
, Vol. 7m No. 2, pp 95-97, 1994. Wendt III has proposed a method which first determines the type of an MR image by reading the image header information, and then computes the display parameters depending on the type of the image. Unfortunately, different rules must be set for different types and orientations of MR images in this method. This makes the algorithm impractical, since new rules need to be added in the program to reflect any new changes in the MR image acquisition process, such as, for example, the use of new coils or new pulse sequences. Ohhashi et al. has developed a neural network based method for the automatic adjustment of the display window. A. Ohhashi, S. Yamada, K Haruki, H. Hatano, Y Fujii, K Yamaguchi and H Ogata, “Automatic adjustment of display window for MR images using a neural network”,
Proceeding of SPIE
, Vol. 1444
, Omage Capture, Formatting and Display
, pp. 63-74, 1991. This method is still a pure histogram based method, and as such, there is a potential problem of very different adjustments for images with very different spatial distributions but very similar histograms. In addition, this method only uses a single neural network for approximating the human adjustment, which is too complicated for a wide range of MR images to be sufficiently approximated with good generalization power by a single neural network.
Recently, the inventors proposed a comprehensive hierarchical neural networks (HNN) based algorithm for automatic and robust adjustment of the display window, which is the subject of U.S. patent application Ser. No. 08/885,080 entitled “Robust and Automatic Adjustment of Display Window Width and Center for MR Images” filed on Jun. 30, 1997, now U.S. Pat. No. 5,995,644, the entire disclosure of which is incorporated herein by reference. This algorithm is based on the principle of learning from examples, (i.e. a large set of MR images associated with the window width/center values adjusted by human experts). This HNN based algorithm uses both wavelet histogram features and spatial statistical information of MR images for feature generation, which overcomes the problem of using pure histogram information only. A hierarchical neural network was developed to decompose the very complicated function approximation problem into several simple subproblems. The hierarchical neural Rtworks are comprised of a modified competitive layer neural network for clustering any input image into a certain number of clusters, and the Radial Basis Function (RBF) and the Bi-modal Linear Estimation (BLE) networks for each class to provide good estimation results. Finally, a data fusion step is used to intelligently combine the multiple estimates from the RBF and BLE networks to provide accurate and robust estimation results.
All the above methods lack the capabilities of adapting the window width/center adjustment to different personal preferences or different viewing conditions as described above. The automatic display window width and center adjustment by using all the previous methods can only be optimized for a particular personal taste and for a particular viewing condition. The demand for an adaptive and automatic windowing system is very common since the automatic windowing system needs to be used by many persons with different personal preferences and in different viewing conditions.
SUMMARY OF THE INVENTION
The present invention is based on the inventors previously described HNN based display window width/center estimation algorithm, the entire disclosure of which is incorporated herein by reference. Due to the basic learning-from-examples principle of the HNN algorithm, the training data set is organized into categories as well as sequences with some representative images selected from each sequence. By re-adjusting some representative images of the sequences in some categories selected by the user, the width/center values of the entire training data are mapped through a global mapping function and a sequence mapping process to adapt the width/center settings of the training data set to the user's preference and the viewing conditions. The global mapping function used is a linear spline function that is obtained by fitting the data of all the re-adjusted representative images. This is used to capture the general personal preference and account for the effect of the viewing condition. The sequence mapping process involves fitting the baseline width/center values in a sequence with low-order polynomial functions and setting the new width/center values in one sequence with the new baseline polynomial function. The sequence mapping is applied after the global mapping and only to the sequence with representative frames selected by the user for re-adjustment. After the above mapping process, the hierarchical neural networks are re-trained with the new training data, thus the re-trained HNN algorithm is adapted to the user's personal preference as well as the viewing condition.
REFERENCES:
patent: 5305204 (1994-04-01), Ohhashi
patent: 5995644 (1999-12-01), Lai et al.
Ohhashi et al. “Application of a Neural Network to Automatic Gray-level Adjustment for Medical Images.” IEEE International Joint Conference on Neural Networks, IJCNN '91. Cat No. 91CH3065-0. pp. 974-980, Nov. 1991.
Fang Ming
Lai Shang-Hong
Mehta Bhavesh
Paschburg Donald B.
Siemens Corporate Research Inc.
LandOfFree
Neural network based auto-windowing system for MR images does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Neural network based auto-windowing system for MR images, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Neural network based auto-windowing system for MR images will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2438762