RATIONALE AND OBJECTIVES: Rotator cuff disorders are prevalent and can cause pain and reduced range of motion and strength. Accurate, noninvasive diagnosis of rotator cuff disorders is therefore important. In this work, we study the relationship between several three-dimensional (3D) shape measurements of the supraspinatus and its pathologic conditions. The objective is to explore the utility of 3D shape descriptors in distinguishing supraspinatus pathologies, leading to computer-aided diagnosis of rotator cuff disorders. MATERIALS AND METHODS: We acquired magnetic resonance images of the shoulder from 73 patients, separated into five pathology groups: normal (14), tear (20), tear and atrophy (13), tear and retraction (15), and tear and atrophy and retraction (11). We segmented the 3D surface of the supraspinatus from each magnetic resonance image, and computed 11 3D shape characteristics for each. We performed an analysis of variance (ANOVA) test for each measurement to test the null hypothesis that the means of the pathology groups were equal. The most promising of the measurements, as determined by the ANOVA test, were used to train a support vector machine classifier to automatically assign new supraspinata to the correct pathology groups. RESULTS: The ANOVA test results rejected the null hypothesis (p < .0045) for 7 of our 11 measurements. Highlights of the results from the support vector machine classifier were 79% accuracy in distinguishing normals from abnormals, and 82% accuracy in distinguishing atrophy from retraction, our main clinical motivation. These scores were tabulated based on leave-one-out cross-validation. CONCLUSION: From the results, we draw the conclusion that 3D shape analysis may be helpful in the diagnosis of rotator cuff disorders, but further investigation is required to develop a 3D shape descriptor that yields ideal pathology group separation. The results of this study suggest several promising avenues of future research to meet this goal.
RATIONALE AND OBJECTIVES:Rotator cuff disorders are prevalent and can cause pain and reduced range of motion and strength. Accurate, noninvasive diagnosis of rotator cuff disorders is therefore important. In this work, we study the relationship between several three-dimensional (3D) shape measurements of the supraspinatus and its pathologic conditions. The objective is to explore the utility of 3D shape descriptors in distinguishing supraspinatus pathologies, leading to computer-aided diagnosis of rotator cuff disorders. MATERIALS AND METHODS: We acquired magnetic resonance images of the shoulder from 73 patients, separated into five pathology groups: normal (14), tear (20), tear and atrophy (13), tear and retraction (15), and tear and atrophy and retraction (11). We segmented the 3D surface of the supraspinatus from each magnetic resonance image, and computed 11 3D shape characteristics for each. We performed an analysis of variance (ANOVA) test for each measurement to test the null hypothesis that the means of the pathology groups were equal. The most promising of the measurements, as determined by the ANOVA test, were used to train a support vector machine classifier to automatically assign new supraspinata to the correct pathology groups. RESULTS: The ANOVA test results rejected the null hypothesis (p < .0045) for 7 of our 11 measurements. Highlights of the results from the support vector machine classifier were 79% accuracy in distinguishing normals from abnormals, and 82% accuracy in distinguishing atrophy from retraction, our main clinical motivation. These scores were tabulated based on leave-one-out cross-validation. CONCLUSION: From the results, we draw the conclusion that 3D shape analysis may be helpful in the diagnosis of rotator cuff disorders, but further investigation is required to develop a 3D shape descriptor that yields ideal pathology group separation. The results of this study suggest several promising avenues of future research to meet this goal.
Authors: Jennifer Krcmery; Rajesh Gupta; Rudyard W Sadleir; Molly J Ahrens; Sol Misener; Christine Kamide; Philip Fitchev; Douglas W Losordo; Susan E Crawford; Hans-Georg Simon Journal: PLoS One Date: 2013-11-20 Impact factor: 3.240