Tendinopathy discrimination using spatial frequency parameters and artificial neural networks

Pengfei Song, Kristopher R. Linstrom, A. John Boye, Kornelia Kulig, Judith Burnfield, Gregory R Bashford

Research output: Chapter in Book/Report/Conference proceedingConference contribution

2 Citations (Scopus)

Abstract

Healthy tendon's structural characteristics are related to the anisotropic speckle pattern observed in ultrasound images. This speckle orientation is disrupted upon damage to the tendon structure as observed in patients with tendinopathy. Quantification of the structural appearance of tendon shows promise in creating a tool for diagnosing, prognosing, or monitoring changes in tendon organization over time. Previously we showed the feasibility of using spatial frequency parameters and Linear Discriminant Analysis (LDA) to categorize tendon tissue as normal or tendinopathic with better than 80% accuracy. The current work aimed to improve accuracy by developing an Artificial Neural Network (ANN) classifier and compare its results with those achieved in our previous LDA work. The eight spatial frequency parameters used in our previous work were extracted from regions of interest (ROI) on tendon images, filtered and classified using an ANN classifier. The spatial frequency parameters were used as inputs to the ANN. For a tendon tested with an ANN trained for that type of tendon, the accuracy was very high, with a correct classification rate (CCR) of 95-99%. However, when testing a tendon with an ANN trained for a different tendon type, the CCR was only 72-75%. This seems to indicate that a unique ANN needs to be trained for each type of tendon. The high CCRs obtained using the tendon-specific ANNs suggest that this novel discrimination strategy may lead to a robust tool for diagnosing and monitoring a degenerated tendon's response to treatment. The ANN CCR was higher than the highest CCR of 82.6% obtained by the LDA used in our previous work.

Original languageEnglish (US)
Title of host publication2009 IEEE International Ultrasonics Symposium and Short Courses, IUS 2009
DOIs
StatePublished - Dec 1 2009
Event2009 IEEE International Ultrasonics Symposium, IUS 2009 - Rome, Italy
Duration: Sep 20 2009Sep 23 2009

Publication series

NameProceedings - IEEE Ultrasonics Symposium
ISSN (Print)1051-0117

Conference

Conference2009 IEEE International Ultrasonics Symposium, IUS 2009
CountryItaly
CityRome
Period9/20/099/23/09

Fingerprint

tendons
discrimination
classifiers
speckle patterns

Keywords

  • Artificial neural network
  • Classifier
  • Image analysis
  • Tendinopathy
  • Tendon
  • Ultrasound

ASJC Scopus subject areas

  • Acoustics and Ultrasonics

Cite this

Song, P., Linstrom, K. R., Boye, A. J., Kulig, K., Burnfield, J., & Bashford, G. R. (2009). Tendinopathy discrimination using spatial frequency parameters and artificial neural networks. In 2009 IEEE International Ultrasonics Symposium and Short Courses, IUS 2009 [5441973] (Proceedings - IEEE Ultrasonics Symposium). https://doi.org/10.1109/ULTSYM.2009.5441973

Tendinopathy discrimination using spatial frequency parameters and artificial neural networks. / Song, Pengfei; Linstrom, Kristopher R.; Boye, A. John; Kulig, Kornelia; Burnfield, Judith; Bashford, Gregory R.

2009 IEEE International Ultrasonics Symposium and Short Courses, IUS 2009. 2009. 5441973 (Proceedings - IEEE Ultrasonics Symposium).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Song, P, Linstrom, KR, Boye, AJ, Kulig, K, Burnfield, J & Bashford, GR 2009, Tendinopathy discrimination using spatial frequency parameters and artificial neural networks. in 2009 IEEE International Ultrasonics Symposium and Short Courses, IUS 2009., 5441973, Proceedings - IEEE Ultrasonics Symposium, 2009 IEEE International Ultrasonics Symposium, IUS 2009, Rome, Italy, 9/20/09. https://doi.org/10.1109/ULTSYM.2009.5441973
Song P, Linstrom KR, Boye AJ, Kulig K, Burnfield J, Bashford GR. Tendinopathy discrimination using spatial frequency parameters and artificial neural networks. In 2009 IEEE International Ultrasonics Symposium and Short Courses, IUS 2009. 2009. 5441973. (Proceedings - IEEE Ultrasonics Symposium). https://doi.org/10.1109/ULTSYM.2009.5441973
Song, Pengfei ; Linstrom, Kristopher R. ; Boye, A. John ; Kulig, Kornelia ; Burnfield, Judith ; Bashford, Gregory R. / Tendinopathy discrimination using spatial frequency parameters and artificial neural networks. 2009 IEEE International Ultrasonics Symposium and Short Courses, IUS 2009. 2009. (Proceedings - IEEE Ultrasonics Symposium).
@inproceedings{4d17589afc3242f78f6a774d339edf03,
title = "Tendinopathy discrimination using spatial frequency parameters and artificial neural networks",
abstract = "Healthy tendon's structural characteristics are related to the anisotropic speckle pattern observed in ultrasound images. This speckle orientation is disrupted upon damage to the tendon structure as observed in patients with tendinopathy. Quantification of the structural appearance of tendon shows promise in creating a tool for diagnosing, prognosing, or monitoring changes in tendon organization over time. Previously we showed the feasibility of using spatial frequency parameters and Linear Discriminant Analysis (LDA) to categorize tendon tissue as normal or tendinopathic with better than 80{\%} accuracy. The current work aimed to improve accuracy by developing an Artificial Neural Network (ANN) classifier and compare its results with those achieved in our previous LDA work. The eight spatial frequency parameters used in our previous work were extracted from regions of interest (ROI) on tendon images, filtered and classified using an ANN classifier. The spatial frequency parameters were used as inputs to the ANN. For a tendon tested with an ANN trained for that type of tendon, the accuracy was very high, with a correct classification rate (CCR) of 95-99{\%}. However, when testing a tendon with an ANN trained for a different tendon type, the CCR was only 72-75{\%}. This seems to indicate that a unique ANN needs to be trained for each type of tendon. The high CCRs obtained using the tendon-specific ANNs suggest that this novel discrimination strategy may lead to a robust tool for diagnosing and monitoring a degenerated tendon's response to treatment. The ANN CCR was higher than the highest CCR of 82.6{\%} obtained by the LDA used in our previous work.",
keywords = "Artificial neural network, Classifier, Image analysis, Tendinopathy, Tendon, Ultrasound",
author = "Pengfei Song and Linstrom, {Kristopher R.} and Boye, {A. John} and Kornelia Kulig and Judith Burnfield and Bashford, {Gregory R}",
year = "2009",
month = "12",
day = "1",
doi = "10.1109/ULTSYM.2009.5441973",
language = "English (US)",
isbn = "9781424443895",
series = "Proceedings - IEEE Ultrasonics Symposium",
booktitle = "2009 IEEE International Ultrasonics Symposium and Short Courses, IUS 2009",

}

TY - GEN

T1 - Tendinopathy discrimination using spatial frequency parameters and artificial neural networks

AU - Song, Pengfei

AU - Linstrom, Kristopher R.

AU - Boye, A. John

AU - Kulig, Kornelia

AU - Burnfield, Judith

AU - Bashford, Gregory R

PY - 2009/12/1

Y1 - 2009/12/1

N2 - Healthy tendon's structural characteristics are related to the anisotropic speckle pattern observed in ultrasound images. This speckle orientation is disrupted upon damage to the tendon structure as observed in patients with tendinopathy. Quantification of the structural appearance of tendon shows promise in creating a tool for diagnosing, prognosing, or monitoring changes in tendon organization over time. Previously we showed the feasibility of using spatial frequency parameters and Linear Discriminant Analysis (LDA) to categorize tendon tissue as normal or tendinopathic with better than 80% accuracy. The current work aimed to improve accuracy by developing an Artificial Neural Network (ANN) classifier and compare its results with those achieved in our previous LDA work. The eight spatial frequency parameters used in our previous work were extracted from regions of interest (ROI) on tendon images, filtered and classified using an ANN classifier. The spatial frequency parameters were used as inputs to the ANN. For a tendon tested with an ANN trained for that type of tendon, the accuracy was very high, with a correct classification rate (CCR) of 95-99%. However, when testing a tendon with an ANN trained for a different tendon type, the CCR was only 72-75%. This seems to indicate that a unique ANN needs to be trained for each type of tendon. The high CCRs obtained using the tendon-specific ANNs suggest that this novel discrimination strategy may lead to a robust tool for diagnosing and monitoring a degenerated tendon's response to treatment. The ANN CCR was higher than the highest CCR of 82.6% obtained by the LDA used in our previous work.

AB - Healthy tendon's structural characteristics are related to the anisotropic speckle pattern observed in ultrasound images. This speckle orientation is disrupted upon damage to the tendon structure as observed in patients with tendinopathy. Quantification of the structural appearance of tendon shows promise in creating a tool for diagnosing, prognosing, or monitoring changes in tendon organization over time. Previously we showed the feasibility of using spatial frequency parameters and Linear Discriminant Analysis (LDA) to categorize tendon tissue as normal or tendinopathic with better than 80% accuracy. The current work aimed to improve accuracy by developing an Artificial Neural Network (ANN) classifier and compare its results with those achieved in our previous LDA work. The eight spatial frequency parameters used in our previous work were extracted from regions of interest (ROI) on tendon images, filtered and classified using an ANN classifier. The spatial frequency parameters were used as inputs to the ANN. For a tendon tested with an ANN trained for that type of tendon, the accuracy was very high, with a correct classification rate (CCR) of 95-99%. However, when testing a tendon with an ANN trained for a different tendon type, the CCR was only 72-75%. This seems to indicate that a unique ANN needs to be trained for each type of tendon. The high CCRs obtained using the tendon-specific ANNs suggest that this novel discrimination strategy may lead to a robust tool for diagnosing and monitoring a degenerated tendon's response to treatment. The ANN CCR was higher than the highest CCR of 82.6% obtained by the LDA used in our previous work.

KW - Artificial neural network

KW - Classifier

KW - Image analysis

KW - Tendinopathy

KW - Tendon

KW - Ultrasound

UR - http://www.scopus.com/inward/record.url?scp=77952836361&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=77952836361&partnerID=8YFLogxK

U2 - 10.1109/ULTSYM.2009.5441973

DO - 10.1109/ULTSYM.2009.5441973

M3 - Conference contribution

SN - 9781424443895

T3 - Proceedings - IEEE Ultrasonics Symposium

BT - 2009 IEEE International Ultrasonics Symposium and Short Courses, IUS 2009

ER -