Automatic Bone Segmentation in Pelvis Area with Bone Marrow Metastases; Applications in Breast Cancer Treatment Monitoring


avatar Faezeh Sanaei Nezhad 1 , avatar Pedram Fadavi 2 , avatar Mohsen Shojaie Moghadam 3 , avatar Hamid Soltanian Zadeh 4 , avatar Hamidreza Saligheh Rad 5 , *

College of Engineering, University of Tehran
Radiation Oncology Department, Iran University of Medical Sciences
Medical Imaging Center, Payambaran Hospital
Radiology Department, Henry Ford Health System, Detroit, Michigan, USA
Tehran University of Medical Sciences, Tehran, IR Iran

how to cite: Sanaei Nezhad F, Fadavi P, Shojaie Moghadam M, Soltanian Zadeh H, Saligheh Rad H. Automatic Bone Segmentation in Pelvis Area with Bone Marrow Metastases; Applications in Breast Cancer Treatment Monitoring. Innov J Radiol. 2014;11(30th Iranian Congress of Radiology):e21334.



In bone marrow metastatic cancers, therapy goals are delay skeletal-related events, reduce symptoms, improve patients life quality and increase their survival. Correct knowledge about patients bone state in shortest possible time is the key to achieve therapy goals. Therefore, several quantitative analyses of MR images have been used, in which accurate segmentation of the bone plays a vital role. Due to heterogeneous nature of tumors and presence of cystic or necrotic areas, correct separation of the overall border of bone is difficult, creating source of error for the quantitative analysis results. Diffusion-weighted (DW) MR images are capable of revealing information of bone status, showing applications in skeletal related diseases. The poor anatomy available in these images gears them toward anatomical MR images such as T1-wighted (T1-w) images. However, boundary distortion created by bone metastasis hierarchically affects information extracted from DW-MR images. Commonly used bone segmentation methods are applicable on healthy bone structures, while they partly fail in presence of bone metastasis, especially in advanced situations.


In this study, we proposed an automatic method to extract pelvis bone from T1-w images in the presence of bone marrow metastasis. This method eliminates manual segmentation variability, resulting reproducible quantitative analysis independent of human error. We also investigated clinical quantitative analyses variations caused by miss-segmentation of the bone due to the bone marrow metastasis, and how our proposed method eliminates such errors.

Patients and Methods:

Pelvis images of ten female patients, with metastatic bone marrow breast cancer under treatment, were carried out on 1.5T MR with the following specifications: TR/TE = 171/4.76 ms, FOV = 430 430, slice thickness = 5 mm, spaces between slices = 5.5 mm. Whole-body DW images acquired using GE-EPI sequence with the following specifications: TR/TE = 5540/102 ms, percent phase FOV =430430, slice thickness=5 mm, spaces between slices = 5.5 mm, at b-value = 50, 900 sec/mm2. ADC-maps were then calculated from DW images. Image processing and analysis: Image processing consisted of the following steps: 1) automatic non-rigid registration of T1w MR images with the ADC-maps with SPM8 software; 2) segmentation of bone in each T1-w image using local level-set with prior information algorithm, implemented in MATLAB; 3) using result of step 2 as a mask, for bone extraction from corresponding DW-MR images such as ADC-map. In order to show the applicability of the automatic segmentation-based analysis, its performance in ADC-map analysis was compared to manual-based analysis methods.


An example of bone structure was illustrated which was diminished due to bone marrow metastasis. Registration and segmentation procedure leads to bone structure extraction regardless of the disease, providing an efficient and accurate bone study. In order to exhibit the variation of information extracted from DW-MR images based on ROI selection, three common parameters extracted from ADC-map based on manual segmentation (performed by 10 different unbiased observers) against parameters extracted based on proposed automatic segmentation method, suggesting significant differences in analysis results. Automatic segmentation accuracy has been evaluated using boundary-based (Hausdorff parameter, result: 5.55 + 1.4) and volume-based (Dice, result: 83% + 0.06 and sensitivity parameter, result: 81% + 0.08) measures. Performance of proposed method was quantitatively evaluated, comparing its performance with two commonly used different methods, in terms of segmentation accuracy. The proposed method is superior to the other two methods with Dice values of 76% and 67%, respectively.


In conclusion, the proposed automatic algorithm has promising results in bone segmentation and in the presence of bone marrow metastasis. This method is accurate and robust, which improves the design of the patients treatment plan and provides the opportunity for accurate high profile analyses. This segmentation is a significant step towards accurate information extraction and therefore, applicable usage of DW-MRI images in bone marrow metastatic cancers, bringing consistency and reproducibility to the extracted information.

Full Text

Full text is available in PDF

© 2014, Author(s). This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License ( which permits copy and redistribute the material just in noncommercial usages, provided the original work is properly cited.