Typically, fluoroscopy includes a plurality of image intensifiers or two-dimensional, flat panel radiation detectors which convert X-ray radiation traversing a patient examination area into electronic signals. Each radiation detector includes a radiation sensitive face, such as a scintillation crystal, which converts the received radiation into a corresponding quantity of light. Solid state diodes are often provided to convert the light emitted by the scintillation crystal into analog electrical signals indicative of the intensity of the crystal emitted light, hence the intensity of the received radiation. The analog signals are converted into corresponding digital signals which are reconstructed into digital images.

Unfortunately, many flat panel radiation detectors, especially large-area flat panel detectors, contain single pixel defects, line defects, double-line defects, and column defects which lead to the generation of defective digital image representations. Prior art techniques correct such pixel, line, and column defects with a series of processes, typically consisting of defect map correction and median filtering. In defect map correction techniques, a base defect map of each panel detector is created during the manufacture of the flat panel detector. Additional defect maps may be created during subsequent calibrations of the panel detectors. These defect

maps are used for the first order detection of permanent defects in the panels and interpolations, such as a median filter, are used to correct these permanent defects. A median filter algorithm is also applied to the entire image in order to provide secondary defect correction for random defects that do not have fixed patterns. This multi-phase defect correction process suffers from processing complexity and inefficiency.

Conventional median filter algorithms adjust all of the pixels of an image representation. In other words, each pixel of an image is replaced by the median value of the pixels in the neighborhood of the particular pixel being examined.

This type of median filtering blurs images, which results in great reduction in image resolution. Further, conventional median filters cannot correct double line and double column defects. In fact, conventional median filters can introduce additional image defects by incorrectly replacing pixels adjacent to defective lines and columns with median values of neighborhood pixels.

The present invention contemplates a new and improved method for detecting and correcting digital image defects which overcomes the above-referenced problems and others.

Summary of the Invention In accordance with one aspect of the present invention, a method for real-time detection and correction of digital image defects due to defective detector pixels includes identifying a candidate pixel value in image data, which includes a plurality of candidate pixel values and corresponding kernels of neighboring pixel values. For each candidate pixel value, a reference value is calculated from the neighboring pixel values. The method further includes comparing a relationship between the candidate pixel value and the reference value with a threshold criterion. Based on the comparison, either the candidate pixel value is replaced with a function of neighboring pixel values or the candidate pixel value is retained.

In accordance with another aspect of the present invention, an adaptive median filter for detecting and

correcting defects in a digital image representation due to defective detectors in a radiation detector array includes a memory which stores each candidate pixel value and neighboring pixel values. A processor calculates a reference value from the neighboring pixel values and compares a relationship between the candidate pixel value and the reference value with a threshold criterion. Based on the comparison, either the candidate pixel value is replaced with a function of the neighboring pixel values or the candidate pixel value is retained.

One advantage of the present invention is that it simplifies the detection and correction of defects in images acquired using flat panel radiation detectors.

Another advantage of the present invention is that it corrects image data dynamically on the fly without a priori mapping or calibration.

Another advantage of the present invention is that it corrects image defects without reducing overall image resolution.

Another advantage of the present invention is that it corrects double line and double column defects.

Another advantage of the present invention is that it corrects image defects without creating additional defects.

Yet another advantage of the present invention resides in its combining image defect detection and correction into a single procedure.

Still another advantage of the present invention is that it leaves most image data unaltered.

Other benefits and advantages of the present invention will become apparent to those skilled in the art upon a reading and understanding of the preferred embodiments.

Brief Description of the Drawings The invention may take form in various components and arrangements of components, and in various steps and arrangements of steps. The drawings are only for purposes of

illustrating preferred embodiments and are not to be construed as limiting the invention.

FIGURE 1 is a diagrammatic illustration of a Fluoro Assistant CT system (FACTs) attached to a CT scanner employing the adaptive median filter in accordance with the present invention; and, FIGURE 2 is a flow chart illustrating details of the defect detection and correction procedure in accordance with the present invention.

Detailed Description of the Preferred Embodiments With reference to FIGURE 1, a fluoroscopic system 10 radiographically examines and generates diagnostic images of a subject disposed on a patient support 12. More specifically, a volume of interest of the subject on the support 12 is moved into an examination region 14. An x-ray tube 16 mounted on a rotating gantry projects a beam of radiation through the examination region 14. A collimator 18 collimates the beam of radiation in one dimension.

The two-dimensional x-ray detectors 20 includes a two-dimensional array of photodetectors connected or preferably integrated into an integrated circuit. A scintillator, comprising a thallium-doped CsI layer, is deposited directly on the photodetector array. X-rays that have traversed the examination region 14 are received through the front face of the scintillation crystal. The scintillation crystal converts these x-rays into a flash or scintillation of visible light of a characteristic wavelength. The visible light exits the scintillation layer via a surface that is optically coupled to the photodetectors. Light from the scintillation layer is converted by the photodetector into corresponding electrical signals indicative of the intensity of the received radiation which is indicative of the integrated x-ray absorption along the corresponding ray between the x-ray rube and the scintillation layer segment.

The electrical signals, along with information on the angular position of the rotating gantry, are digitized by analog-to-digital converters. The digital diagnostic data is processed for offset and gain calibration by an image calibration processor 30. The digital image representation includes a rectangular array of digital pixel values, each indicating the gray scale of a corresponding image pixel. For simplicity of illustration, a two-dimensional array corresponding to a projection image is described in detail.

However, it is to be appreciated that the present technique is also applicable to three-dimensional arrays representing a volume.

When the projection image representation is generated, lines of pixel values are passed through an adaptive filter 32, preferably a median filter. The adaptive median filter 32 performs a real-time detection and correction of image defects. Such image defects may be due to pixel defects, line defects, double-line defects, column defects, and double- column defects in the two-dimensional detector panel 20, as well as random defects. For an nxn adaptive filter, read out lines of pixel values are temporarily stored in n-1 digital line memory devices 341, 342/.... In the illustrated 3x3 embodiment, the buffer stores the two preceding lines. A field programmable gate array (FPGA) 40 reads the current and two preceding data lines. As the oldest data line is read out of one buffer, the current data line is read into it.

The FPGA 40 includes a comparitor circuit or processor 42 which compares the pixel values of the three lines with threshold criteria 44. Various threshold criteria are contemplated. Preferably, each pixel value of the middle line is compared with the eight immediately surrounding pixel values in itself and in the two adjoining data lines. If a pixel value varies by 20% or another preselected percentage from the median value of its eight nearest neighbors, an adaptive filter processor 46 replaces it with the median value of its nearest neighbors. Rather than (or in addition to) the 20% threshold

criteria, each pixel value can be compared with other criteria including full black and full white. The adaptive filter replaces each pixel value that fails these criteria with a median or other preselected function of its nearest neighbors that are not full black or white. Pixel values which pass the test are not altered by the adaptive filter. In this manner, any (if any) pixel values of the middle data line that failed the test are replaced with median filtered values and are passed by the adaptive filter for further processing. After the middle data line is scanned with the nxn kernel, the lines of data are indexed with a new line added and the most remote line dropped.

The implementation of the adaptive median filter in a pipelined architecture yields one processed pixel output for every unprocessed input pixel, often referred to as a systolic processor. The outputs are delayed with respect to the input by the pipeline processing delay time. More specifically, after nxn pixels are latched by the FPGA, a sorting algorithm within the FPGA yields the median value of the nxn kernel. In parallel, the unprocessed value of the given pixel is stored and made available along with the median value. Also in parallel, a multiplier within the FPGA computes a threshold value for the pixel being examined by multiplying the unprocessed or original pixel value by the predefined defect threshold, 0.2 for example. The difference between the original unprocessed pixel value and the median value is determined by a sort and subtraction algorithm within the FPGA and then compared to the threshold value. If the difference value is greater than the threshold value, the median value is substituted for the original pixel value at the output of the FPGA. Otherwise, the original pixel value is at the output of the FPGA. Artisans will appreciate that all data, including original pixel, median value, threshold value, and difference value, are synchronized through pipelined latches. Further, the horizontal and vertical raster synchronization signal timing relationship with respect to a given pixel is also maintained using shift registers.

The filtered image is stored in a volumetric image memory 50. A video processor 52 processes the defect-corrected image to create projection images, and reformats them for display on a monitor 54, such as a video or LCD monitor.

With reference to FIGURE 2 and continuing reference to FIGURE 1, a more detailed method and software based apparatus for detecting and correcting digital image defects begins at step 100 with the inputting of lines of digital pixel values and a predefined defect threshold into the adaptive median filter. Again, the predefined defect threshold is used by the adaptive median filter to determine whether a given pixel of the image should be replaced by the median value of the neighboring pixels or should be left unaltered. The inputted line of pixel values is then copied 110 into a correction memory for processing. A kernel of nxn pixels is selected 120, with the central pixel value of the kernel being the candidate pixel value to be examined. In one embodiment, a 3x3 kernel is selected with the center pixel of the kernel being examined and compared to the eight adjacent nearest neighbor pixels.

The selected nxn kernel is reordered 130. More particularly, the pixel values of the selected nxn kernel are sorted numerically by value and adjacent pixels of like value are merged into a single pixel value. A median value of the reordered and condensed kernel is calculated 140. For example, in a 3x3 kernel of nine pixels, the pixel value that is being processed is compared to the median value of the nine pixels in the kernel. However, before the median value of the kernel is calculated, any pixel values in the kernel of like value are combined or condensed into a single representation of the common value. For example, three adjacent pixels may each have a value of"1". These three pixels are then merged into a single merged pixel having a value of"1". Then, a median value of the six pixel values reordered kernel is calculated.

The median is advantageous for its computational simplicity, speed and ability to correct double line and double column

defects. However, other functions of the surrounding pixel values based on spread, slope, weighted averages, more complex and other functions are also contemplated.

A threshold value for the particular pixel being examined is calculated 150. The threshold value is calculated, in the preferred embodiment, by multiplying the candidate pixel value by the predetermined defect threshold criteria. In addition, a reference value is calculated 160 by subtracting the median value of the selected kernel from the candidate pixel value. The calculated difference value 160 is then compared 170 to the calculated threshold value 150. If the difference value is greater than the threshold value, the original pixel value is replaced 180 by the median value of the kernel in which the candidate pixel is located 180. If the difference value is less than the threshold value, the candidate pixel is determined to be free of defect and the original candidate pixel value remains unchanged.

In one embodiment, the defect threshold is selected to be 20%. In other words, candidate pixel values which differ by greater than 20% from the median value of the kernel in which they are located are replaced by the median value of the kernel. Conversely, candidate pixel values that are within 20% of the median value of the kernel in which they are located remain at their original unprocessed pixel value. In an alternate embodiment, the defect threshold may be chosen such that the adaptive median filter searches only for pixels having a zero value, i. e. dark, or a maximum value, i. e. white. It is to be appreciated that the following filtering procedure is performed on all of the pixel values for real-time detection and correction of image defects.

It is to be appreciated that the adaptive median filter is effective in correcting pixel defects, line and column defects, including double line and double column defects, bipolar line and column defects, such as one line white and adjacent line black, cluster pixel defects, ASIC boundary lines, driver line noises, noisy pixels, and the like.

Although median filters are particularly effective, it is contemplated that other filters and interpolation techniques can be utilized. For example, the pixel being examined and any zero level pixels and saturated pixels can be excluded from the kernel median. Pixel values in the kernel can be preferentially weighted. Other kernels, such as larger kernels, non-square kernels, and the like are also contemplated. Higher order interpolations may also be utilized.