The impact of simulated motion blur on lesion detection performance in full-field digital mammography

Abstract

Objective: Motion blur is a known phenomenon in full-field digital mammography, but the impact on lesion detection is unknown. This is the first study to investigate detection performance with varying magnitudes of simulated motion blur.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 blurring (0 mm) and two magnitudes of simulated blurring (0.7 mm and 1.5 mm).Abnormal cases were biopsy proven. Mathematical simulation was used to provide a pixel shift in order to simulate motion blur. A free-response observer study was conducted to compare lesion detection performance for the three conditions. The equally weighted jackknife alternative freeresponse receiver operating characteristic (wJAFROC) was used as the figure of merit. Test alpha was set at 0.05 to control probability of Type I error.Results: wJAFROC analysis found a statistically significant difference in lesion detection Advances in knowledge:This research demonstrates for the first time that motion blur has a negative and statistically significant impact on lesion detection performance digital mammography.

Automatic detection of simulated motion blur in mammograms

“…Moreover, Ma et al investigated the minimum amount of simulated motion blur detected visually by radiologists, which was determined to be 0.7 mm. The realism of their mathematical blur simulation model was validated by an observer study and was found to be comparable to real blur in mammograms.…”

“…Although limited work has been done to quantify the effects of motion blur on radiologists' performance, there is evidence that motion blur might not be detected visually by a human observer and it can negatively affect lesion detection performance. 1,3,7 As of this date, no other has study investigated the ability of machine-learning classifiers and BM operators to detect motion blur in mammograms; this work has. When distinguishing blurry from unblurry mammograms, the average classification accuracies were above 85% and 92% for the INbreast and DDSM mammographic databases, respectively, when using Ensemble Bagged Trees, fine Gaussian SVM, and weighted KNN.…”

“…Breast tissue movement was modeled as the movement of a single particle moving under random motion. 3,7,13,14 The random motion blur approach adopted in this study is based on the simulated motion blur model discussed in Ma et al 3 and Boracchi and Foi. 13,14 The random motion is generated by allowing the particle to follow a twodimensional (2D) trajectory in the continuous domain, where a complex-valued vector determines the discrete positions of the particle.…”

“…Previous studies show the paddle motion can be up to 1.5 mm in the vertical direction during a mammography exam, 6,18 and radiologists were capable of visually detecting simulated blur at 1.5 mm of motion. 3,7 Therefore, the choice of the blur PSF mask size was limited to 1.5 mm of pixel displacement to represent 1.5 mm of breast tissue motion, focusing on simulating lower blur levels that radiologists often fail to detect. 3 Figure 1 shows sample motion trajectories and the blur PSF masks obtained from them when sampled on a discrete grid.…”

“…For the same amount of simulated motion blur, application of the simulated-blur in this study may not deliver the same level of visual blur compared to the blur model in Ma et al 3 Another limitation of the motion-blur simulation is that image noise may be blurred by the blur simulation, in contrast to reality where real motion blur does not affect quantum mottle noise. 7 The model could possibly be adjusted by mathematically simulating quantum mottle noise and re-introducing it to the mammograms post-blurring.…”