Digital Image Preprocessing and Processing (Rescaling)

Chapter 29: Digital Image Preprocessing and Processing (Rescaling)

Objectives

Upon completion of this chapter, you should be able to:

  1. Define preprocessing and postprocessing.
  2. Describe five types of corrections made to the acquired image to establish field uniformity.
  3. Explain how spatial kernels can be used to correct for del drop-out.
  4. Define partitioned pattern recognition and exposure field recognition.
  5. Explain the construction of the image histogram.
  6. Give two examples of how the computer can identify key landmarks within the image histogram.
  7. Define the characteristics of three general types of histogram analysis, and why they must be matched to the actual acquired histogram.
  8. Define the values of interest.
  9. Interpret the gray scale curve.
  10. Explain how bizarre histogram shapes can cause errors in rescaling.
  11. Describe how pixel values in the image are re-mapped to rescale brightness and correct for scatter radiation.
  12. Define the limitations on what the computer can do with the dataset from a latent image.

INTRODUCTION

The term "post-processing" is commonly associated with the processing option available for the user and is often distinguished from the default processing that all digital radiographs are subjected to. In reality, this distinction is arbitrary since available processing options are generally identical. Ideally, the default processing should be chosen so that no additional "postprocessing" is necessary, as discussed by Prokop et al. (2003).

Throughout this book, the capability for postprocessing has been highlighted as a major advantage of digital imaging over conventional imaging, primarily defined as the ability to manipulate the image multiple times without re-exposing the patient to radiation. Conversely, preprocessing refers to the computerized operations applied to "raw" digital images to correct flaws in image acquisition due to the elements and circuitry of the image receptor system or processor.

Definitions
  • Preprocessing: All corrections made to the "raw" digital image data due to physical flaws in image acquisition inherent to specific image receptor systems (also termed acquisition processing).
  • Postprocessing: All manipulation and adjustments of the digital image after corrections for data acquisition have been made, targeting refinement of the image, which may or may not be part of the default processing.

This functional delineation clarifies that preprocessing focuses on image corrections, while postprocessing pertains to image refinement. However, the term "processing" is best used synonymously with rescaling of the image, as detailed in subsequent sections.

PREPROCESSING I: FIELD UNIFORMITY

Digital imaging systems, particularly direct-capture DR systems, face inherent limitations in image acquisition that necessitate field uniformity corrections. Direct-capture DR systems experience electronic faults affecting individual detector elements (dels) leading to noise or pixel loss. Pre-display processes compensate for these issues, involving additional stages of correction for noise and dark current exposure.

Dark Noise and Flat-Field Corrections
  • Dark noise: Refers to accumulated background exposure to a CR phosphor plate and dark current effects.
  • Flat-field corrections: A uniform signal correction across the imaging field. This is performed using a low exposure to the receptor without objects, comparing pixel values in central and corner regions.

PREPROCESSING II: NOISE REDUCTION FOR DEL DROP-OUT

Del drop-out involves complete failure of individual dels due to issues like TFT failure. This data loss is corrected through mathematical interactivity using kernels, which average surrounding pixels to fill dead spaces. This spatial processing method can repair mild drop-out by interpolation but necessitates detection plate replacement for severe cases.

Spatial Kernels
  • A kernel is a sub-matrix passed over the image matrix to execute mathematical functions. An averaging kernel fills in missing data based on surrounding pixel values and is particularly useful in correcting del drop-out.

PREPROCESSING III: IMAGE ANALYSIS

Partitioned Pattern Recognition and Exposure Field Recognition

Image analysis entails segmentation, exposure field recognition, and histogram construction and analysis.

  • Segmentation: Identifies edges of exposed fields to ensure accurate histogram construction.
  • Exposure field recognition (EFR): Essential for detecting background densities that could skew histogram data.
Histogram Construction

An image histogram counts the number of pixels at each density value, presented in a bar graph format. Key features include:

  • Values of Interest (VOI): Range of pixel values to be processed during image manipulation, crucial in enhancing specific anatomical structures.
  • Histogram Analysis: Determine critical landmarks (SMIN, SMAX) necessary for processing and eliminating unwanted data in construction.

RESCALING (PROCESSING) THE IMAGE

Rescaling adjusts pixel values so images appear "normal". The process may involve algebraic functions utilizing look-up tables (LUT). The ideal LUT aligns with the system's evident dynamic range, and algorithms ensure brightness and gray scale adjustments are effectively administered.

Effect of Histogram Shapes

Bizarre shapes can lead to histogram analysis errors affecting the outcome of processing. Proper procedures should exclude background densities, ensuring accurate rescaling without distortions.

Processing Steps Overview
  • Field uniformity corrections: Address image acquisition flaws and smooth brightness.
  • Noise and del drop-out corrections: Minimize discrepancies in pixel data leading to errors.
  • Histogram construction and analysis: Organize pixel data effectively for processing.
  • Rescaling: Utilize LUT to adjust images appropriately for display.

SUMMARY AND REVIEW QUESTIONS

The chapter concludes with a review of key concepts, outlining the distinctions between preprocessing and postprocessing, the significance of histogram analysis, and the nuances of image rescaling.

Review Questions
  1. Why does DR require more preprocessing than CR?
  2. Describe the mathematical process for correcting del drop-out using kernels.
  3. Explain the importance of selecting the appropriate type of histogram analysis corresponding to the acquired histogram data.