Automated segmentation of pelvic bone structure in x-ray radiographs using active shape models and directed Hough transform

Abstract

Traumatic pelvic injuries are often associated with severe, life-threatening hemorrhage, and immediate medical treatment is therefore vital. However, patient prognosis depends heavily on the type, location and severity of the bone fracture. Detecting these fractures can prove challenging for physicians due to the complexity of the pelvic bone structure, but is vital in identifying the most severe cases. X-ray imaging is fast and requires disruption to the patient, and is an important initial diagnostic step upon a patient's admission. Therefore, automated fracture detection from initial patient x-rays can assist physicians in diagonsis and treatment. However, fractures in different areas of the pelvis display different characteristics in x-ray images. As a first and crucial step, the separate structures within the pelvisin particular, the ilium and left and right acetabulum -must therefore be correctly segmented and identified in the x-ray image. This paper describes a algorithm for segmentation of these structures, using a hierarchical approach that combines directed Hough transform and Active Shape models. Unlike various other x-ray segmentation methods, the algorithm presented here both handles the inconsistencies between x-ray images in a clinical environment, and performs successfully in the presence of fracture.

Traumatic Pelvic Injury Outcome Prediction by Extracting Features from Relevant Medical Records and X-Ray Images

“…Segmentation of pelvic S-ray images into distinct structures was already performed in our previous studies [7], [8]. The method used here is a Combined Spline/ASM algorithm, which improves the standard Active Shape Model (ASM) approach by using spline interpolation to regulate deformation.…”

“…The success of ASM is also highly dependent on correct initialization, and if the starting shape is placed incorrectly, the algorithm cannot converge to the correct edges. This study therefore uses the automatic initialization method described in [7] to ensure correct object detection.…”