Computerized Quantification of Pain Drawings

Søren O'Neill*, Tue Secher Jensen, Peter Kent

*Corresponding author for this work

Research output: Contribution to journalJournal articleResearchpeer-review

Abstract

Using a computer algorithm to quantify pain drawings could be useful, especially when large numbers of drawings need to be assessed. Whilst informal visual assessment of pain drawings can give clinicians a quick impression of the extent of pain and its location, formal quantification of pain drawings by computer for research purposes is not necessarily trivial. The current study compared seven different approaches to quantification in a large sample of clinical spinal pain drawings. A large number (n = 55,720) of pain drawings were extracted from the SpineData database, a clinical registry of spinal pain patients in the Region of Southern Denmark. Drawings were analyzed both as pixel (raster) and vector based images, with different approaches based on the raw pain drawing, simple encircling polygons, convex-hull encircling polygons and discrete anatomical regions. Data were analyzed using principal component analysis, correlation and linear regression, as well as informal visual inspection of outlier pain drawings. Eighty-one percent of the variance could be explained by the first principal component, which we interpreted as the true score variance, i.e. the variance attributable to differences in pain area between individuals. The second principal component explained 10% of the variance and was loaded differentially by polygon-based methods and non-polygon-based methods. Correlations between the different approaches ranged from 0.66 to 1.00. Some approaches correlated so strongly as to be interchangeable, others tended to bias area estimates significantly. Visual inspection of outlier pain drawing indicated that when the different approaches to quantification yielded different results, characteristic patterns could be identified in the style and patterns of those pain drawings. The different approaches reflected the same underlying construct (pain area), but could not be relied upon to produce the same area estimates and were affected by the interaction between drawing style and quantification approach. To some extend, the "correct" choice of quantification method is specific to and dictated by the style of each pain drawing. A differentiated approach is required in which the results of quantification and the drawing style are considered in combination. We provide suggestions for such differentiated approaches taking into account the nature of the drawing data (raster vs. vector) and the method of analysis (partly vs completely automated). The chosen method of quantifying pain drawings in combination with the drawing style of the individual patient, can impact the resulting area estimate to a significant degree. These issues should be considered before undertaking computerized area estimation of pain drawings.

Original languageEnglish
JournalScandinavian Journal of Pain
ISSN1877-8879
DOIs
Publication statusE-pub ahead of print - 12. Oct 2019

Fingerprint

Pain Measurement
Denmark
Registries
Linear Models
Databases
Research

Keywords

  • back pain
  • computer analysis
  • pain drawing
  • pain measure
  • quantification
  • spinal pain

Cite this

@article{761167aac9fc49fa870471c434f7a323,
title = "Computerized Quantification of Pain Drawings",
abstract = "Using a computer algorithm to quantify pain drawings could be useful, especially when large numbers of drawings need to be assessed. Whilst informal visual assessment of pain drawings can give clinicians a quick impression of the extent of pain and its location, formal quantification of pain drawings by computer for research purposes is not necessarily trivial. The current study compared seven different approaches to quantification in a large sample of clinical spinal pain drawings. A large number (n = 55,720) of pain drawings were extracted from the SpineData database, a clinical registry of spinal pain patients in the Region of Southern Denmark. Drawings were analyzed both as pixel (raster) and vector based images, with different approaches based on the raw pain drawing, simple encircling polygons, convex-hull encircling polygons and discrete anatomical regions. Data were analyzed using principal component analysis, correlation and linear regression, as well as informal visual inspection of outlier pain drawings. Eighty-one percent of the variance could be explained by the first principal component, which we interpreted as the true score variance, i.e. the variance attributable to differences in pain area between individuals. The second principal component explained 10{\%} of the variance and was loaded differentially by polygon-based methods and non-polygon-based methods. Correlations between the different approaches ranged from 0.66 to 1.00. Some approaches correlated so strongly as to be interchangeable, others tended to bias area estimates significantly. Visual inspection of outlier pain drawing indicated that when the different approaches to quantification yielded different results, characteristic patterns could be identified in the style and patterns of those pain drawings. The different approaches reflected the same underlying construct (pain area), but could not be relied upon to produce the same area estimates and were affected by the interaction between drawing style and quantification approach. To some extend, the {"}correct{"} choice of quantification method is specific to and dictated by the style of each pain drawing. A differentiated approach is required in which the results of quantification and the drawing style are considered in combination. We provide suggestions for such differentiated approaches taking into account the nature of the drawing data (raster vs. vector) and the method of analysis (partly vs completely automated). The chosen method of quantifying pain drawings in combination with the drawing style of the individual patient, can impact the resulting area estimate to a significant degree. These issues should be considered before undertaking computerized area estimation of pain drawings.",
keywords = "back pain, computer analysis, pain drawing, pain measure, quantification, spinal pain, back pain, computer analysis, pain drawing, pain measure, quantification, spinal pain",
author = "S{\o}ren O'Neill and Jensen, {Tue Secher} and Peter Kent",
year = "2019",
month = "10",
day = "12",
doi = "10.1515/sjpain-2019-0082",
language = "English",
journal = "Scandinavian Journal of Pain",
issn = "1877-8860",
publisher = "De Gruyter",

}

Computerized Quantification of Pain Drawings. / O'Neill, Søren; Jensen, Tue Secher; Kent, Peter.

In: Scandinavian Journal of Pain, 12.10.2019.

Research output: Contribution to journalJournal articleResearchpeer-review

TY - JOUR

T1 - Computerized Quantification of Pain Drawings

AU - O'Neill, Søren

AU - Jensen, Tue Secher

AU - Kent, Peter

PY - 2019/10/12

Y1 - 2019/10/12

N2 - Using a computer algorithm to quantify pain drawings could be useful, especially when large numbers of drawings need to be assessed. Whilst informal visual assessment of pain drawings can give clinicians a quick impression of the extent of pain and its location, formal quantification of pain drawings by computer for research purposes is not necessarily trivial. The current study compared seven different approaches to quantification in a large sample of clinical spinal pain drawings. A large number (n = 55,720) of pain drawings were extracted from the SpineData database, a clinical registry of spinal pain patients in the Region of Southern Denmark. Drawings were analyzed both as pixel (raster) and vector based images, with different approaches based on the raw pain drawing, simple encircling polygons, convex-hull encircling polygons and discrete anatomical regions. Data were analyzed using principal component analysis, correlation and linear regression, as well as informal visual inspection of outlier pain drawings. Eighty-one percent of the variance could be explained by the first principal component, which we interpreted as the true score variance, i.e. the variance attributable to differences in pain area between individuals. The second principal component explained 10% of the variance and was loaded differentially by polygon-based methods and non-polygon-based methods. Correlations between the different approaches ranged from 0.66 to 1.00. Some approaches correlated so strongly as to be interchangeable, others tended to bias area estimates significantly. Visual inspection of outlier pain drawing indicated that when the different approaches to quantification yielded different results, characteristic patterns could be identified in the style and patterns of those pain drawings. The different approaches reflected the same underlying construct (pain area), but could not be relied upon to produce the same area estimates and were affected by the interaction between drawing style and quantification approach. To some extend, the "correct" choice of quantification method is specific to and dictated by the style of each pain drawing. A differentiated approach is required in which the results of quantification and the drawing style are considered in combination. We provide suggestions for such differentiated approaches taking into account the nature of the drawing data (raster vs. vector) and the method of analysis (partly vs completely automated). The chosen method of quantifying pain drawings in combination with the drawing style of the individual patient, can impact the resulting area estimate to a significant degree. These issues should be considered before undertaking computerized area estimation of pain drawings.

AB - Using a computer algorithm to quantify pain drawings could be useful, especially when large numbers of drawings need to be assessed. Whilst informal visual assessment of pain drawings can give clinicians a quick impression of the extent of pain and its location, formal quantification of pain drawings by computer for research purposes is not necessarily trivial. The current study compared seven different approaches to quantification in a large sample of clinical spinal pain drawings. A large number (n = 55,720) of pain drawings were extracted from the SpineData database, a clinical registry of spinal pain patients in the Region of Southern Denmark. Drawings were analyzed both as pixel (raster) and vector based images, with different approaches based on the raw pain drawing, simple encircling polygons, convex-hull encircling polygons and discrete anatomical regions. Data were analyzed using principal component analysis, correlation and linear regression, as well as informal visual inspection of outlier pain drawings. Eighty-one percent of the variance could be explained by the first principal component, which we interpreted as the true score variance, i.e. the variance attributable to differences in pain area between individuals. The second principal component explained 10% of the variance and was loaded differentially by polygon-based methods and non-polygon-based methods. Correlations between the different approaches ranged from 0.66 to 1.00. Some approaches correlated so strongly as to be interchangeable, others tended to bias area estimates significantly. Visual inspection of outlier pain drawing indicated that when the different approaches to quantification yielded different results, characteristic patterns could be identified in the style and patterns of those pain drawings. The different approaches reflected the same underlying construct (pain area), but could not be relied upon to produce the same area estimates and were affected by the interaction between drawing style and quantification approach. To some extend, the "correct" choice of quantification method is specific to and dictated by the style of each pain drawing. A differentiated approach is required in which the results of quantification and the drawing style are considered in combination. We provide suggestions for such differentiated approaches taking into account the nature of the drawing data (raster vs. vector) and the method of analysis (partly vs completely automated). The chosen method of quantifying pain drawings in combination with the drawing style of the individual patient, can impact the resulting area estimate to a significant degree. These issues should be considered before undertaking computerized area estimation of pain drawings.

KW - back pain

KW - computer analysis

KW - pain drawing

KW - pain measure

KW - quantification

KW - spinal pain

KW - back pain

KW - computer analysis

KW - pain drawing

KW - pain measure

KW - quantification

KW - spinal pain

U2 - 10.1515/sjpain-2019-0082

DO - 10.1515/sjpain-2019-0082

M3 - Journal article

JO - Scandinavian Journal of Pain

JF - Scandinavian Journal of Pain

SN - 1877-8860

ER -