Bhavya R. Shah MD, Presenter: Nothing to Disclose

Jiajing Xu MS, Abstract Co-Author: Nothing to Disclose

Bao H. Do MD, Abstract Co-Author: Nothing to Disclose

Daniel L. Rubin MD, Abstract Co-Author: Grant, General Electric Company

Sandy Napel PhD, Abstract Co-Author: Medical Advisory Board, Fovia, Inc Consultant, Carestream Health, Inc

Christopher Frederick Beaulieu MD, PhD, Abstract Co-Author: Nothing to Disclose

Accurate diagnosis of focal bone lesions on radiography is challenging due to overlapping features in aggressive and nonaggressive processes. This project aims to develop a system to enable automated diagnostic decision support of focal bone lesions by utilizing quantitative radiographic image feature analysis.

The system utilizes a user-drawn region of interest (ROI) and computes statistical features from a comprehensive panel of over 18 feature sets. A control experiment was designed by placing a ROI within a lesion and another ROI in the same contralateral extremity in an area of visually normal bone. 10 enchondromas and 10 aneurysmal bone cysts were analyzed on 600 dpi resolution radiographs. The lesion was circumscribed by a radiologist to delimit the lesion in each case and the images were processed to extract quantitative image features across 18 feature sets. The mean values and s.d. of each feature within each lesion type (ABC and enchondroma) were calculated, and analyzed for significance with a T-test. Because their visual features varied, enchondromas were subsequently subdivided into those in large or small bones. Statistical significance between ABC vs. large enchondroma (LE), ABC vs. small enchondroma (SE), and LE vs. SE groups was calculated.

Control experiments showed higher histogram mean density comparing fibrous dysplasia with normal bone (13.4%) and in comparing an enchondroma with normal bone (7.8%). The skewness of the histogram also varied between lesions and the surrounding bone, changing 24% for fibrous dysplasia and 1064% for the enchondroma. Edge Sharpness feature statistics were significantly different between the LE and ABC populations (p range 0.005 to 0.044). 18 feature statistics were significant between the LE and SE groups (p range<0.0001 to 0.049), including 17 edge sharpness and 1 histogram skewness calculations.

We developed a semi-automated image-processing system to analyze two skeletal tumor types. The preliminary results suggest statistically significant differences in features between lesions. With further systematic analysis, the system offers an objective, scalable, and potentially useful means of diagnosing bone diseases

Our system is broadly applicable in terms of imaging modalities (e.g. x-ray, CT, and MRI), disease states (metabolic, infectious, and neoplastic conditions, etc.), and tasks such as content based ima

Shah, B, Xu, J, Do, B, Rubin, D, Napel, S, Beaulieu, C, Quantitative Analysis of Focal Bone Tumors on Radiographs: Initial Experience Comparing Lucent and Sclerotic Lesions.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12035387.html