Abstract
Evaluation of segmentation algorithms usually involves comparisons of segmentations to gold-standard delineations without regard to the ultimate medical decision-making task. We compare two segmentation evaluations methods - a Dice similarity coefficient (DSC) evaluation and a diagnostic classification task-based evaluation method using lesions from breast computed tomography. In our investigation, we use results from two previously developed lesion-segmentation algorithms [a global active contour model (GAC) and a global with local aspects active contour model]. Although similar DSC values were obtained (0.80 versus 0.77), we show that the global + local active contour (GLAC) model, as compared with the GAC model, is able to yield significantly improved classification performance in terms of area under the receivers operating characteristic (ROC) curve in the task of distinguishing malignant from benign lesions. [Area under the ROC curve (AUC)=0.78 compared to 0.63, p蠐0.001]. This is mainly because the GLAC model yields better detailed information required in the calculation of morphological features. Based on our findings, we conclude that the DSC metric alone is not sufficient for evaluating segmentation lesions in computer-aided diagnosis tasks.
| Original language | English (US) |
|---|---|
| Article number | 031012 |
| Journal | Journal of Medical Imaging |
| Volume | 1 |
| Issue number | 3 |
| DOIs | |
| State | Published - Oct 1 2014 |
Fingerprint
Keywords
- breast computed tomography
- breast mass classification
- computer-aided diagnosis/detection
- image analysis
- segmentation
ASJC Scopus subject areas
- Bioengineering
- Radiology Nuclear Medicine and imaging
Cite this
Impact of lesion segmentation metrics on computer-aided diagnosis/detection in breast computed tomography. / Kuo, Hsien Chi; Giger, Maryellen L.; Reiser, Ingrid; Drukker, Karen; Boone, John M; Lindfors, Karen K; Yang, Kai; Edwards, Alexandra.
In: Journal of Medical Imaging, Vol. 1, No. 3, 031012, 01.10.2014.Research output: Contribution to journal › Article
}
TY - JOUR
T1 - Impact of lesion segmentation metrics on computer-aided diagnosis/detection in breast computed tomography
AU - Kuo, Hsien Chi
AU - Giger, Maryellen L.
AU - Reiser, Ingrid
AU - Drukker, Karen
AU - Boone, John M
AU - Lindfors, Karen K
AU - Yang, Kai
AU - Edwards, Alexandra
PY - 2014/10/1
Y1 - 2014/10/1
N2 - Evaluation of segmentation algorithms usually involves comparisons of segmentations to gold-standard delineations without regard to the ultimate medical decision-making task. We compare two segmentation evaluations methods - a Dice similarity coefficient (DSC) evaluation and a diagnostic classification task-based evaluation method using lesions from breast computed tomography. In our investigation, we use results from two previously developed lesion-segmentation algorithms [a global active contour model (GAC) and a global with local aspects active contour model]. Although similar DSC values were obtained (0.80 versus 0.77), we show that the global + local active contour (GLAC) model, as compared with the GAC model, is able to yield significantly improved classification performance in terms of area under the receivers operating characteristic (ROC) curve in the task of distinguishing malignant from benign lesions. [Area under the ROC curve (AUC)=0.78 compared to 0.63, p蠐0.001]. This is mainly because the GLAC model yields better detailed information required in the calculation of morphological features. Based on our findings, we conclude that the DSC metric alone is not sufficient for evaluating segmentation lesions in computer-aided diagnosis tasks.
AB - Evaluation of segmentation algorithms usually involves comparisons of segmentations to gold-standard delineations without regard to the ultimate medical decision-making task. We compare two segmentation evaluations methods - a Dice similarity coefficient (DSC) evaluation and a diagnostic classification task-based evaluation method using lesions from breast computed tomography. In our investigation, we use results from two previously developed lesion-segmentation algorithms [a global active contour model (GAC) and a global with local aspects active contour model]. Although similar DSC values were obtained (0.80 versus 0.77), we show that the global + local active contour (GLAC) model, as compared with the GAC model, is able to yield significantly improved classification performance in terms of area under the receivers operating characteristic (ROC) curve in the task of distinguishing malignant from benign lesions. [Area under the ROC curve (AUC)=0.78 compared to 0.63, p蠐0.001]. This is mainly because the GLAC model yields better detailed information required in the calculation of morphological features. Based on our findings, we conclude that the DSC metric alone is not sufficient for evaluating segmentation lesions in computer-aided diagnosis tasks.
KW - breast computed tomography
KW - breast mass classification
KW - computer-aided diagnosis/detection
KW - image analysis
KW - segmentation
UR - http://www.scopus.com/inward/record.url?scp=85019281896&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85019281896&partnerID=8YFLogxK
U2 - 10.1117/1.JMI.1.3.031012
DO - 10.1117/1.JMI.1.3.031012
M3 - Article
AN - SCOPUS:85019281896
VL - 1
JO - Journal of Medical Imaging
JF - Journal of Medical Imaging
SN - 0720-048X
IS - 3
M1 - 031012
ER -