David R Holmes PhD, Presenter: Nothing to Disclose

Rickey Carter PhD, Abstract Co-Author: Nothing to Disclose

Kurt Ernest Augustine MS, Abstract Co-Author: Nothing to Disclose

Yu Liu MD, Abstract Co-Author: Nothing to Disclose

Maria Shiung, Abstract Co-Author: Nothing to Disclose

Lifeng Yu PhD, Abstract Co-Author: Nothing to Disclose

Phillip Edwards, Abstract Co-Author: Nothing to Disclose

Cynthia H. McCollough PhD, Abstract Co-Author: Research Grant, Siemens AG

Joel Garland Fletcher MD, Abstract Co-Author: Grant, Siemens AG

While numerous CT noise reduction methods have been developed, it is difficult to directly measure the clinical impact of each approach. We have developed an open source computer workstation to efficiently conduct observer studies of low dose CT protocols to determine the superiority or non-inferiority of new reconstruction methods.

The workstation allows a user to conduct lesion detection and characterization, and image quality assessment in a time-efficient manner. The user is required to identify the location and size of all lesions in a dataset by delineating the long axis of the lesion. Both manual and automatic software tools have been developed to match corresponding lesions between an observer and routine dose FBP reference standard. The automatic matching algorithm computes correspondence by determining if the reference ROI overlaps with an observer ROI. Matching rules are employed to insure lesions are appropriately characterized (e.g., benign/malignant) if they are detected. The algorithm reports true positives (TP), false positives (FP), and false negatives (FN) to a back-end database for export and JAFROC analysis.

The automated matching algorithm was validated using ten radiologist observers – each reviewing 10 datasets. The study PI created the reference standard based on correlative imaging, follow-up and pathology reports. Observers required an average of 5.6 minutes (range 0.5 – 25.4) min to review each case. The PI completed semi-automated visual matching of observer and reference marks and diagnoses. The observers delineated a combined 644 lesions (including TP, FP, and FN) across all 10 observers. Automated matching required < 1 second and correctly matched 94.7% of the lesions (compared to the manual matching). Incorrect responses by the algorithm included 11 overmatched (e.g. multiple overlapping ROIs) detections and 23 mis-matches between reference and observer ROIs.

A system for interactively evaluating CT denoising methods must minimize radiologist effort, accurately match reference detections and classifications with observer markings using automated and manual visual tools, and create a streamlined workflow and statistical analysis.

Dedicated workstations for observer performance in low dose CT minimize radiologist effort with streamlined workflow and provide automated and visual tools for reference standard matching.

Holmes, D, Carter, R, Augustine, K, Liu, Y, Shiung, M, Yu, L, Edwards, P, McCollough, C, Fletcher, J, Development of a Dedicated Workstation to Facilitate Rapid Performance of Observer Studies in Low-dose CT.  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13013150.html