Active relearning for robust supervised training of emphysema patterns

Sushravya Raghunath, Srinivasan Rajagopalan, Ronald A. Karwoski, Brian Jack Bartholmai, Richard A. Robb

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

Radiologists are adept at recognizing the character and extent of lung parenchymal abnormalities in computed tomography (CT) scans. However, the inconsistent differential diagnosis due to subjective aggregation necessitates the exploration of automated classification based on supervised or unsupervised learning. The robustness of supervised learning depends on the training samples. Towards optimizing emphysema classification, we introduce a physician-in-the-loop feedback approach to minimize ambiguity in the selected training samples. An experienced thoracic radiologist selected 412 regions of interest (ROIs) across 15 datasets to represent 124, 129, 139 and 20 training samples of mild, moderate, severe emphysema and normal appearance, respectively. Using multi-view (multiple metrics to capture complementary features) inductive learning, an ensemble of seven un-optimized support vector models (SVM) each based on a specific metric was constructed in less than 6 s. The training samples were classified using seven SVM models and consensus labels were created using majority voting. In the active relearning phase, the ensemble-expert label conflicts were resolved by the expert. The efficacy and generality of active relearning feedback was assessed in the optimized parameter space of six general purpose classifiers across the seven dissimilarity metrics. The proposed just-in-time active relearning feedback with un-optimized SVMs yielded 15 % increase in classification accuracy and 25 % reduction in the number of support vectors. The average improvement in accuracy of six classifiers in their optimized parameter space was 21 %. The proposed cooperative feedback method enhances the quality of training samples used to construct automated classification of emphysematous CT scans. Such an approach could lead to substantial improvement in quantification of emphysema.

Original languageEnglish (US)
Pages (from-to)548-555
Number of pages8
JournalJournal of Digital Imaging
Volume27
Issue number4
DOIs
StatePublished - 2014

Fingerprint

Emphysema
Feedback
Supervised learning
Learning
Tomography
Labels
Classifiers
Unsupervised learning
Politics
Consensus
Differential Diagnosis
Thorax
Agglomeration
Physicians
Lung
Radiologists

Keywords

  • Active relearning
  • Emphysema
  • Supervised classification
  • Support Vector Machines (SVM)
  • Training sample cleaning

ASJC Scopus subject areas

  • Radiology Nuclear Medicine and imaging
  • Radiological and Ultrasound Technology
  • Computer Science Applications

Cite this

Active relearning for robust supervised training of emphysema patterns. / Raghunath, Sushravya; Rajagopalan, Srinivasan; Karwoski, Ronald A.; Bartholmai, Brian Jack; Robb, Richard A.

In: Journal of Digital Imaging, Vol. 27, No. 4, 2014, p. 548-555.

Research output: Contribution to journalArticle

Raghunath, Sushravya ; Rajagopalan, Srinivasan ; Karwoski, Ronald A. ; Bartholmai, Brian Jack ; Robb, Richard A. / Active relearning for robust supervised training of emphysema patterns. In: Journal of Digital Imaging. 2014 ; Vol. 27, No. 4. pp. 548-555.
@article{aec66dcd399f4578995f22b01271c6f5,
title = "Active relearning for robust supervised training of emphysema patterns",
abstract = "Radiologists are adept at recognizing the character and extent of lung parenchymal abnormalities in computed tomography (CT) scans. However, the inconsistent differential diagnosis due to subjective aggregation necessitates the exploration of automated classification based on supervised or unsupervised learning. The robustness of supervised learning depends on the training samples. Towards optimizing emphysema classification, we introduce a physician-in-the-loop feedback approach to minimize ambiguity in the selected training samples. An experienced thoracic radiologist selected 412 regions of interest (ROIs) across 15 datasets to represent 124, 129, 139 and 20 training samples of mild, moderate, severe emphysema and normal appearance, respectively. Using multi-view (multiple metrics to capture complementary features) inductive learning, an ensemble of seven un-optimized support vector models (SVM) each based on a specific metric was constructed in less than 6 s. The training samples were classified using seven SVM models and consensus labels were created using majority voting. In the active relearning phase, the ensemble-expert label conflicts were resolved by the expert. The efficacy and generality of active relearning feedback was assessed in the optimized parameter space of six general purpose classifiers across the seven dissimilarity metrics. The proposed just-in-time active relearning feedback with un-optimized SVMs yielded 15 {\%} increase in classification accuracy and 25 {\%} reduction in the number of support vectors. The average improvement in accuracy of six classifiers in their optimized parameter space was 21 {\%}. The proposed cooperative feedback method enhances the quality of training samples used to construct automated classification of emphysematous CT scans. Such an approach could lead to substantial improvement in quantification of emphysema.",
keywords = "Active relearning, Emphysema, Supervised classification, Support Vector Machines (SVM), Training sample cleaning",
author = "Sushravya Raghunath and Srinivasan Rajagopalan and Karwoski, {Ronald A.} and Bartholmai, {Brian Jack} and Robb, {Richard A.}",
year = "2014",
doi = "10.1007/s10278-014-9686-z",
language = "English (US)",
volume = "27",
pages = "548--555",
journal = "Journal of Digital Imaging",
issn = "0897-1889",
publisher = "Springer New York",
number = "4",

}

TY - JOUR

T1 - Active relearning for robust supervised training of emphysema patterns

AU - Raghunath, Sushravya

AU - Rajagopalan, Srinivasan

AU - Karwoski, Ronald A.

AU - Bartholmai, Brian Jack

AU - Robb, Richard A.

PY - 2014

Y1 - 2014

N2 - Radiologists are adept at recognizing the character and extent of lung parenchymal abnormalities in computed tomography (CT) scans. However, the inconsistent differential diagnosis due to subjective aggregation necessitates the exploration of automated classification based on supervised or unsupervised learning. The robustness of supervised learning depends on the training samples. Towards optimizing emphysema classification, we introduce a physician-in-the-loop feedback approach to minimize ambiguity in the selected training samples. An experienced thoracic radiologist selected 412 regions of interest (ROIs) across 15 datasets to represent 124, 129, 139 and 20 training samples of mild, moderate, severe emphysema and normal appearance, respectively. Using multi-view (multiple metrics to capture complementary features) inductive learning, an ensemble of seven un-optimized support vector models (SVM) each based on a specific metric was constructed in less than 6 s. The training samples were classified using seven SVM models and consensus labels were created using majority voting. In the active relearning phase, the ensemble-expert label conflicts were resolved by the expert. The efficacy and generality of active relearning feedback was assessed in the optimized parameter space of six general purpose classifiers across the seven dissimilarity metrics. The proposed just-in-time active relearning feedback with un-optimized SVMs yielded 15 % increase in classification accuracy and 25 % reduction in the number of support vectors. The average improvement in accuracy of six classifiers in their optimized parameter space was 21 %. The proposed cooperative feedback method enhances the quality of training samples used to construct automated classification of emphysematous CT scans. Such an approach could lead to substantial improvement in quantification of emphysema.

AB - Radiologists are adept at recognizing the character and extent of lung parenchymal abnormalities in computed tomography (CT) scans. However, the inconsistent differential diagnosis due to subjective aggregation necessitates the exploration of automated classification based on supervised or unsupervised learning. The robustness of supervised learning depends on the training samples. Towards optimizing emphysema classification, we introduce a physician-in-the-loop feedback approach to minimize ambiguity in the selected training samples. An experienced thoracic radiologist selected 412 regions of interest (ROIs) across 15 datasets to represent 124, 129, 139 and 20 training samples of mild, moderate, severe emphysema and normal appearance, respectively. Using multi-view (multiple metrics to capture complementary features) inductive learning, an ensemble of seven un-optimized support vector models (SVM) each based on a specific metric was constructed in less than 6 s. The training samples were classified using seven SVM models and consensus labels were created using majority voting. In the active relearning phase, the ensemble-expert label conflicts were resolved by the expert. The efficacy and generality of active relearning feedback was assessed in the optimized parameter space of six general purpose classifiers across the seven dissimilarity metrics. The proposed just-in-time active relearning feedback with un-optimized SVMs yielded 15 % increase in classification accuracy and 25 % reduction in the number of support vectors. The average improvement in accuracy of six classifiers in their optimized parameter space was 21 %. The proposed cooperative feedback method enhances the quality of training samples used to construct automated classification of emphysematous CT scans. Such an approach could lead to substantial improvement in quantification of emphysema.

KW - Active relearning

KW - Emphysema

KW - Supervised classification

KW - Support Vector Machines (SVM)

KW - Training sample cleaning

UR - http://www.scopus.com/inward/record.url?scp=84904401131&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84904401131&partnerID=8YFLogxK

U2 - 10.1007/s10278-014-9686-z

DO - 10.1007/s10278-014-9686-z

M3 - Article

C2 - 24771303

AN - SCOPUS:84904401131

VL - 27

SP - 548

EP - 555

JO - Journal of Digital Imaging

JF - Journal of Digital Imaging

SN - 0897-1889

IS - 4

ER -