W K Ma1, R Aspin2, J Kelly3, S Millington3, P Hogg1. 1. 1 Directorate of Radiography, University of Salford, Salford, UK. 2. 2 Directorate of Computer Science & Software Engineering, School of Computing, Science and Engineering, University of Salford, Salford, UK. 3. 3 Department of Radiography, Countess of Chester Hospital, Chester, UK.
Abstract
OBJECTIVE: Image blurring in mammography can cause significant image degradation and interpretational problems. A potential source is due to paddle movement during image formation. Paddle movement has been shown to be as much as 1.5 mm. No study has yet been performed to determine how much motion would be noticeable visually. The aim of this study is to determine the minimum amount of simulated breast movement at which blurring can be detected visually. METHODS: 25 artefact-free mammogram images were selected. Mathematical simulation software was created to mimic the effect of blurring produced by breast movement during exposure. Motion simulation was imposed to 15 levels, from 0.1 to 1.5 mm stepping through 0.1 mm increments. 15 degraded images and 1 without blurring were de-identified, randomized and assessed on a blinded basis by two clinical experts to determine the presence or absence of blurring. Statistical testing was carried out to determine the consistency between the two observers. RESULTS: The probability of simulated blurred image detection is the highest for the gaussian method and the lowest for soft-edged mask estimation. CONCLUSION: The amount of simulated breast movement at which blurring can be detected visually for gaussian blur, hard-edge mask estimation and soft-edge mask estimation is 0.4, 0.8 and 0.7 mm, respectively. Cohen's kappa for all the levels of simulated blurring is 0.689 (p < 0.05). ADVANCES IN KNOWLEDGE: This research establishes the concept of using probability to represent visual detection of blurring rather than defining a hard cut-off level.
OBJECTIVE: Image blurring in mammography can cause significant image degradation and interpretational problems. A potential source is due to paddle movement during image formation. Paddle movement has been shown to be as much as 1.5 mm. No study has yet been performed to determine how much motion would be noticeable visually. The aim of this study is to determine the minimum amount of simulated breast movement at which blurring can be detected visually. METHODS: 25 artefact-free mammogram images were selected. Mathematical simulation software was created to mimic the effect of blurring produced by breast movement during exposure. Motion simulation was imposed to 15 levels, from 0.1 to 1.5 mm stepping through 0.1 mm increments. 15 degraded images and 1 without blurring were de-identified, randomized and assessed on a blinded basis by two clinical experts to determine the presence or absence of blurring. Statistical testing was carried out to determine the consistency between the two observers. RESULTS: The probability of simulated blurred image detection is the highest for the gaussian method and the lowest for soft-edged mask estimation. CONCLUSION: The amount of simulated breast movement at which blurring can be detected visually for gaussian blur, hard-edge mask estimation and soft-edge mask estimation is 0.4, 0.8 and 0.7 mm, respectively. Cohen's kappa for all the levels of simulated blurring is 0.689 (p < 0.05). ADVANCES IN KNOWLEDGE: This research establishes the concept of using probability to represent visual detection of blurring rather than defining a hard cut-off level.
Authors: Stephen H Taplin; Carolyn M Rutter; Charles Finder; Margaret T Mandelson; Florence Houn; Emily White Journal: AJR Am J Roentgenol Date: 2002-04 Impact factor: 3.959
Authors: Wang Kei Ma; Mark F McEntee; Claire Mercer; Judith Kelly; Sara Millington; Peter Hogg Journal: Br J Radiol Date: 2016-01-07 Impact factor: 3.039
Authors: Ahmed K Abdullah; Judith Kelly; John D Thompson; Claire E Mercer; Rob Aspin; Peter Hogg Journal: Br J Radiol Date: 2017-06-16 Impact factor: 3.039