Joseph Edwin Burns MD, PhD, Abstract Co-Author: Nothing to Disclose

Jianhua Yao PhD, Abstract Co-Author: Royalties, iCAD, Inc

Yasuyuki Pham MS, MD, Abstract Co-Author: Nothing to Disclose

James Stieger BS, Abstract Co-Author: Nothing to Disclose

Ronald M. Summers MD, PhD, Presenter: Royalties, iCAD, Inc Research funded, iCAD, Inc Stockholder, Johnson & Johnson Grant, Viatronix, Inc

To create an automated computer system to detect, localize, and analyze quantitative characteristics of thoracic and lumbar vertebrae with compression fractures on CT images, to help guide patient management.

A computer system was developed to detect and localize vertebrae with compression fractures, evaluate the axial plane spatial distribution of compression injury, and calculate quantitative characteristics of the fractures, in a series of CT studies from 43 patients (70±13 yrs, range 50-96 yrs, 25 females, 28 males). 32 of the patients had reported compression fractures, and 11 did not. 656 vertebrae were evaluated, in which there were a total of 88 vertebrae with compression fractures. The four steps to the methodology are: spine segmentation and partitioning, endplate detection, height distribution computation, and quantitative compression fracture analysis. For each segmented vertebra, the system detects the vertebral body endplates as local intensity maxima. The cranial-caudal (CC), anterior-posterior, and transverse axes of the vertebral body are computed, forming a local coordinate system. A projection along the CC axis between endplates of the same vertebra is located, and separation distance recorded. The vertebral body is partitioned into 17 axially concentric cells and the height of each cell is quantified. The vertebral height distribution and the height ratio relative to adjacent vertebrae were used as features to train and test the computer system. 10-fold cross validation was employed to evaluate the performance.

The sensitivity was 0.88 (77/88) (95% confidence interval [0.81, 0.93]), with a false positive (FP) rate of 0.63 per patient. There was only one FP in the 11 control cases with no fractures. Compression fracture deformities ranged from 4% to 63% in maximum degree, and were found most commonly in the T7 and L1 vertebral bodies. FP detections occurred most often due to Schmorl’s nodes and image artifacts. False negative detections occurred when multiple compression fractures appeared sequentially.

The system can robustly detect, anatomically localize, and generate quantitative statistics for vertebral body compression fractures on CT.

Quantitative fracture assessment has potential to guide patient management and assist ongoing efforts to develop clinically relevant standards for classification of vertebral compression fractures.

Burns, J, Yao, J, Pham, Y, Stieger, J, Summers, R, Automated Detection and Quantitative Analysis of Spinal Vertebral Compression Fractures on CT.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14015896.html