Aarhus University Seal / Aarhus Universitets segl

Jens Randel Nyengaard

The proportionator: unbiased stereological estimation using biased automatic image analysis and non-uniform probability proportional to size sampling

Research output: Contribution to journal/Conference contribution in journal/Contribution to newspaperJournal articleResearchpeer-review


  • Stereological Research Laboratory
  • Institute of Clinical Medicine
  • The Department of Pathology - ÅKH
The proportionator is a novel and radically different approach to sampling with microscopes based on well-known statistical theory (probability proportional to size - PPS sampling). It uses automatic image analysis, with a large range of options, to assign to every field of view in the section a weight proportional to some characteristic of the structure under study. A typical and very simple example, examined here, is the amount of color characteristic for the structure, marked with a stain with known properties. The color may be specific or not. In the recorded list of weights in all fields, the desired number of fields are sampled automatically with probability proportional to the weight and presented to the expert observer. Using any known stereological probe and estimator, the correct count in these fields leads to a simple, unbiased estimate of the total amount of structure in the sections examined, which in turn leads to any of the known stereological estimates, including size distributions and spatial distributions. The unbiasedness is not a function of the assumed relation between the weight and the structure, which is in practice always a biased relation from a stereological (integral geometric) point of view. The efficiency of the proportionator depends, however, directly on this relation to be positive. The sampling and estimation procedure is simulated in sections with characteristics and various kinds of noises in possibly realistic ranges. In all cases examined, the proportionator is 2- to 15-fold more efficient than the common systematic, uniformly random sampling. The simulations also indicate that the lack of a simple predictor of the coefficient of error (CE) due to field-to-field variation is a more severe problem for uniform sampling strategies than anticipated. Because of its entirely different sampling strategy, based on known but non-uniform sampling probabilities, the proportionator for the first time allows the real CE at the section level to be automatically estimated (not just predicted), unbiased - for all estimators and at no extra cost to the user.
Original languageEnglish
JournalComputers in Biology and Medicine
Pages (from-to)313-328
Number of pages16
Publication statusPublished - 2008

    Research areas

  • Automatic image analysis, PPS-sampling, section quantization, simple random, smooth fractionator, specific stains, stereology, systematic uniform random, tissue inhomogeneity

See relations at Aarhus University Citationformats

Download statistics

No data available

ID: 9220748