Min Zhang, Presenter: Nothing to Disclose

Clinton V. Wellnitz MD, Abstract Co-Author: Nothing to Disclose

William Pavlicek PhD, Abstract Co-Author: Nothing to Disclose

Miao He, Abstract Co-Author: Nothing to Disclose

Amy Kiyo Hara MD, Abstract Co-Author: License agreement, General Electric Company Researcher, General Electric Company

Teresa Wu, Abstract Co-Author: Nothing to Disclose

Excessive Z axis coverage with CT imaging can be a source of unnecessary radiation exposure to patients. Using a tracking database that receives the DICOM Dose Structured Report (containing the CTDIvol and DLP) and a RIS HL7 feed (containing the patient height, weight, anterior-posterior/AP dimensions and lateral/LAT dimensions), a prediction model was constructed to automatically monitor the Z axis coverage. This model was incorporated into the tracking database as a web tool to be used for quality reviews of Z axis coverage.

Excessive Z axis coverage provides an opportunity for significant radiation dose reduction. The ability to automate Z axis coverage review for the abdomen pelvis CT Exam will be helpful in educating technologists, particularly if the performing technologist is recorded as part of the data. This approach can be applied to protocol specific measurement of Z axis coverage including CT colonography, urography and other limited scan length studies.

Expanding the upper bound further (i.e. to 7 cm beyond ideal) will increase the likelihood that Z axis coverage was too large. Adding more features will improve the accuracy and may provide a 'suggested' Z axis coverage that will directly aid the technologist in specifying Z axis coverage.

Data from 271 abdomen pelvis CT exams were used to build the model. Ideal Z axis coverages (from 2 cm above the dome of the liver to 2 cm below the ischial tuberosities) were first discretized into 9 classes, and a resampling method was used to redistribute the unbalanced data. Excessive Z axis coverage, defined as above the predicted upper bound of the ideal coverage, was identified using a Random Forest algorithm. 6 features (height, weight, abdomen AP and LAT dimensions , pelvis AP and LAT dimensions ) were used for the prediction, and the model was validated using 10 fold cross-validation to determine the overall accuracy. The results showed 98.15% accuracy was achieved in detection of excessive Z axis coverage above the upper bound, with 5.5 cm average distance to the ideal coverage.

Zhang, M, Wellnitz, C, Pavlicek, W, He, M, Hara, A, Wu, T, Automated Monitoring of Abdomen Pelvis CT Exams for Excessive Z-Axis Coverage.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12033799.html