The impact of simulated motion blur on breast cancer detection performance in full field digital mammography (FFDM)

Abdullah, AK

Abstract

Objective: Full-field Digital Mammography (FFDM) is employed in breast screening for the early detection of breast cancer. High quality, artefact free, diagnostic images are crucial to the accuracy of this process. Unwanted motion during the image acquisition phase and subsequent image blurring is an unfortunate occurrence in some FFDM images. The research detailed in this thesis seeks to understand the impact of motion blur on cancer detection performance in FFDM images using novel software to perform simulation of motion, an observer study to measure the lesion detection performance and physical measures to assess the impact of simulated motion blur on image characteristics of the lesions.
Method: Seven observers (15±5 years’ reporting experience) evaluated 248 cases (62 containing malignant masses, 62 containing malignant microcalcifications and 124 normal cases) for three conditions: no motion blur (0.0 mm) and two magnitudes of simulated motion blur (0.7 mm and 1.5 mm). Abnormal cases were biopsy proven. A free-response observer study was conducted to compare lesion detection performance for the three conditions. Equally weighted jackknife alternative free-response receiver operating characteristic (wJAFROC) was used as the figure of merit. A secondary analysis of data was deemed important to simulate ‘double reporting’. In this secondary analysis, six of the observers are combined with the seventh observer to evaluate the impact of combined free-response data for lesion detection and to assess if combined two observers data could reduce the impact of simulated motion blur on detection performance. To compliment this, the physical characteristics of the lesions were obtained under the three conditions in order to assess any change in characteristics of the lesions when blur is present in the image. The impact of simulated motion blur on physical characteristics of malignant masses was assessed using a conspicuity index; for microcalcifications, a new novel metric, known as dispersion index, was used.
Results: wJAFROC analysis found a statistically significant difference in lesion detection performance for both masses (F (2,22) = 6.01, P=0.0084) and microcalcifications (F(2,49) = 23.14, P<0.0001). For both lesion types, the figure of merit reduced as the magnitude of simulated motion blur increased. Statistical differences were found between some of the pairs investigated for the detection of masses (0.0mm v 0.7mm, and 0.0mm v 1.5mm) and all pairs for microcalcifications (0.0 mm v 0.7 mm, 0.0 mm v 1.5 mm, and 0.7 mm v 1.5 mm). No difference was detected between 0.7 mm and 1.5 mm for masses.
For combined two observers’ data of masses, there was no statistically significant difference between single and combined free-response data for masses (F(1,6) = 4.04, p=0.1001, -0.031 (-0.070, 0.008) [treatment difference (95% CI)]. For combined data of microcalcifications, there was a statistically significant difference between single and combined free-response data (F(1,6) = 12.28, p=0.0122, -0.056 (-0.095, -0.017) [treatment difference (95% CI)]. Regarding the physical measures of masses, conspicuity index increases as the magnitude of simulated motion blur increases. Statistically significant differences were demonstrated for 0.0–0.7 mm t(22)=-6.158 (p<0.000); 0.0–1.5 mm t(22)=-6.273 (p<0.000); and 0.7–1.5 mm (t(22)=-6.231 (p<0.000). Lesion edge angle decreases as the magnitude of simulated motion blur increases. Statistically significant differences were demonstrated for 0.0–0.7 mm t(22)=3.232 (p<0.004); for 0.0–1.5 mm t(22)=6.592 (p<0.000); and 0.7–1.5mm t(22)=2.234 (p<0.036). For the grey level change there was no statistically significant difference as simulated motion blur increases to 0.7 and then to 1.5mm. For image noise there was a statistically significant difference, where noise reduced as simulated motion blur increased: 0.0–0.7 mm t(22)=22.95 (p<0.000); 0.0–1.5mm t(22)=24.66 (p<0.000); 0.7–1.5 mm t(22)=18.11 (p<0.000). For microcalcifications, simulated motion blur had a negative impact on the ‘dispersion index’.
Conclusion: Mathematical simulations of motion blur resulted in a statistically significant reduction in lesion detection performance. This reduction in performance could have implications for clinical practice. Simulated motion blur has a negative impact on the edge angle of breast masses and a negative impact on the image characteristics of microcalcifications. These changes in the image lesion characteristics appear to have a negative effect on the visual identification of breast cancer.

Citation

Abdullah, A. (in press). The impact of simulated motion blur on breast cancer detection performance in full field digital mammography (FFDM). (Thesis). University of Salford

• Thesis Type: Thesis
• Acceptance Date: Jun 8, 2018
• Deposit Date: Sep 21, 2018
• Publicly Available Date: Sep 21, 2018
• Additional Information: Funders : Ministry of Higher Education and Scientific Research in Iraq;University of Diyala