Development of an automated detection algorithm for patient motion blur in digital mammograms

Abstract Background Patient motion is a common problem during cardiac PET. The purpose of the present study was to investigate to what extent motions influence the quantitative accuracy of cardiac 15O-water PET/CT and to develop a method for automated motion detection. Method Frequency and magnitude of motion was assessed visually using data from 50 clinical 15O-water PET/CT scans. Simulations of 4 types of motions with amplitude of 5 to 20 mm were performed based on data from 10 scans. An automated motion detection algorithm was evaluated on clinical and simulated motion data. MBF and PTF of all simulated scans were compared to the original scan used as reference. Results Patient motion was detected in 68% of clinical cases by visual inspection. All observed motions were small with amplitudes less than half the LV wall thickness. A clear pattern of motion influence was seen in the simulations with a decrease of myocardial blood flow (MBF) in the region of myocardium to where the motion was directed. The perfusable tissue fraction (PTF) trended in the opposite direction. Global absolute average deviation of MBF was 3.1% ± 1.8% and 7.3% ± 6.3% for motions with maximum amplitudes of 5 and 20 mm, respectively. Automated motion detection showed a sensitivity of 90% for simulated motions ≥ 10 mm but struggled with the smaller (≤ 5 mm) simulated (sensitivity 45%) and clinical motions (accuracy 48%). Conclusion Patient motion can impair the quantitative accuracy of MBF. However, at typically occurring levels of patient motion, effects are similar to or only slightly larger than inter-observer variability, and downstream clinical effects are likely negligible.

Download Full-text

Automated detection and association of surface waves

Annals of Geophysics ◽

10.4401/ag-4208 ◽

1994 ◽

Vol 37 (3) ◽

Author(s):

R. G. North ◽

C. R. D. Woodgold

Keyword(s):

Surface Waves ◽

Group Velocity ◽

Rayleigh Waves ◽

Arrival Time ◽

Narrow Band ◽

Broad Band ◽

Detection Algorithm ◽

Automated Detection ◽

Short Period ◽

Group Velocities

An algorithm for the automatic detection and association of surface waves has been developed and tested over an 18 month interval on broad band data from the Yellowknife array (YKA). The detection algorithm uses a conventional STA/LTA scheme on data that have been narrow band filtered at 20 s periods and a test is then applied to identify dispersion. An average of 9 surface waves are detected daily using this technique. Beamforming is applied to determine the arrival azimuth; at a nonarray station this could be provided by poIarization analysis. The detected surface waves are associated daily with the events located by the short period array at Yellowknife, and later with the events listed in the USGS NEIC Monthly Summaries. Association requires matching both arrival time and azimuth of the Rayleigh waves. Regional calibration of group velocity and azimuth is required. . Large variations in both group velocity and azimuth corrections were found, as an example, signals from events in Fiji Tonga arrive with apparent group velocities of 2.9 3.5 krn/s and azimuths from 5 to + 40 degrees clockwise from true (great circle) azimuth, whereas signals from Kuriles Kamchatka have velocities of 2.4 2.9 km/s and azimuths off by 35 to 0 degrees. After applying the regional corrections, surface waves are considered associated if the arrival time matches to within 0.25 km/s in apparent group velocity and the azimuth is within 30 degrees of the median expected. Over the 18 month period studied, 32% of the automatically detected surface waves were associated with events located by the Yellowknife short period array, and 34% (1591) with NEIC events; there is about 70% overlap between the two sets of events. Had the automatic detections been reported to the USGS, YKA would have ranked second (after LZH) in terms of numbers of associated surface waves for the study period of April 1991 to September 1992.

Download Full-text