Removing spikes caused by quantization noise from high-resolution histograms

Magn Reson Med. 2003 Sep;50(3):649-53. doi: 10.1002/mrm.10561.

Abstract

A novel method is presented for the removal of spikes, caused by the division of two series of integers, from high-resolution histograms. When two series of integers are divided the results take the form of a nonuniform distribution. Such a division is often used in medical imaging, due to the storage of most images as integers. An example of this is the division of the saturated and unsaturated signal intensities to obtain a magnetization transfer ratio. Histograms produced using these methods often contain spikes relating to the nonuniform distribution mentioned above. These spikes can have serious implications for certain histogram characteristics. Most commonly, peak height and location can be seriously distorted by these spikes, which have predictable locations. These spikes can be removed by the addition of uniformly distributed noise to the integer signal intensities before division.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Magnetic Resonance Imaging / methods*
  • Signal Processing, Computer-Assisted*