Current Research

Vertebra Analysis

Shuang Liu

Osteoporosis is a serious and growing public health concern worldwide and is characterized by low bone mineral density and architectural deterioration of bone tissue, leading to bone fragility and susceptibility to fracture. It is estimated that about 75 million people in the United States, Europe and Japan are affected by osteoporosis.

Vertebral compression fractures are common in the elderly, accounting for approximately 1.5 million vertebral compression fractures occur every year in the US, which have the potential to cause significant disability and morbidity, as well as incapacitating back pain for many months.

The purpose of this study was to develop a fully automated framework for vertebra analysis on low-dose chest CT (LDCT), which consists of the following four stages:

1. The individual vertebrae are segmented and labeled with anatomical names based on image intensity profile analysis and the spatial constraints established upon other pre-identified organs and structures including clavicles, ribs, sternum and lungs [2].
2. The cortical closed surface of each segmented vertebral body is obtained by employing progressive surface resolution (PSR) algorithm [1].
3. The compression fracture is detected based on image intensity and texture feature analysis of the segmented cortical surface.
4. The osteoporosis is detected based on bone mineral density assessed in the volume of interest surrounded by the segmented cortical surface.

The vertebra segmentation and labeling was validated with 1270 LDCT scans through visual evaluation and achieved satisfactory performance in 89.9% of the scans [2]. The PSR algorithm was applied to the cortical surface segmentation of 460 vertebral bodies on 46 LDCT images [1]. For the visual evaluation, the algorithm achieved acceptable segmentation for 99.35% vertebral bodies. Quantitative evaluation was performed on 46 vertebral bodies and achieved overall mean Dice coefficient of 0.939 (with max = 0.957, min = 0.906 and variance = 0.011) using manual annotations as the ground truth.

Presentations and Publications

1. Liu, S., Xie, Y. & Reeves, A. P. Automated 3D closed surface segmentation: application to vertebral body segmentation in CT image. J Comput Assist Radiol Surg.,11(5), pp.789-801, 2016.
2. Liu, S., Xie, Y., & Reeves, A. P. Individual bone structure segmentation and labeling from low-dose Chest CT. Proceedings of SPIE Medical Imaging, submitted, March 2017.

List of Current Research Projects