Final Report Summary - MOSAIC (Method of osteo-fracture study through automatic identification and classification: biomechanical analysis of bone trabecular structure)
Project introduction
The broad objective of the MOSAIC project was to improve the accuracy in the assessment of bone fracture risk by expanding our understanding of bone fracture event. Nowadays, the main investigation technique for the prediction of the fracture event is based on the study of bone quantity. Nonetheless, this approach showed in the past to be unreliable, with errors ranging from 20 to 40 %. The study aimed to develop tools for the identification, study and finally prediction of the trabecular fracture zone. To address these wide objectives, the following scheme was applied:
(a) data collection;
(b) development of image processing techniques;
(c) fracture classification.
Data collection
During the two years of the project, a number of in vitro bone datasets were collected. In the first period of the project 60 datasets were collected from https://www.physiomespace.com/(opens in new window) and produced by Laboratorio Tecnologia Medica, Istituto Ortopedico Rizzoli, Bologna, Italy, with the financial support of the European Union (EU) project 'Living human digital library' (LHDL IST-2004-026932). The datasets were micro-computed-tomography (micro-CT) images of trabecular bone specimen extracted from several different locations from the lower limb of two donors. During the second half of the project, new datasets were collected. The researcher spent four month at the Eindhoven University of Technology (TU), in the Biomechanics and Tissue Engineering (BMTE) section of the department of Biomedical Engineering, under the supervision of Professor Bert Van Rietbergen. BMTE laboratory was involved in the acquisition of the meso-level data within the VPHOP (ICT-2008-223865) EU project. 17 new micro-level datasets of trabecular specimen were collected. Ten of them were available also for meso-level analysis. Finally, 10 datasets of vertebrae were collected both at micro and meso-level.
Development of image processing techniques
The project aimed to the identification, study and prediction of the trabecular fracture. A registration procedure was developed for the identification of the trabecular fracture zone from micro-CT images. The scheme was based on the application of a rigid registration applied on the segmented trabecular structure. The application of the automatic scheme allowed the identification and study of the full three-dimensional (3D) distribution of the structure. The automatic scheme was validated on micro-level datasets. After its validation the scheme was applied on 60 specimens and the morphometric characteristics of trabecular fractures were studied.
Fracture classification
The study and comprehension of trabecular failure zone allowed the development of a fracture classifier. Simple classifiers based on linear regression were investigated to discriminate fractured and non-fractured zone and for the identification of most significant features. The investigation of more sophisticated learning machines allowed applying different classifiers on micro-CT images of a single specimen for the identification of its failure zone. The classifier showed to be able to predict the fracture zone of a single specimen with an accuracy of 98 % over 734 Volume of Interest analysed. Preliminary results were presented at the 5th Workshop on Artificial Neural Networks in Pattern Recognition.
Deviation from the original project
The application of the procedure on meso-level images was not straightforward. Within the project, the proposed fracture identification scheme should be applied on meso-level images. This result was not fully achieved. The reason was that the segmentation process based on fixed thresholding failed the identification of the trabecular structure in the meso-level datasets. Thus, for the half of the second year, the researcher team was concentrated on the development of a new segmentation technique for the meso-level datasets. The technique developed is based on level-set segmentation and initial results were obtained at the end of the project.
Results and potential impact
During the MOSAIC project, a system for the identification, study and prediction of trabecular failure zone in micro-CT images was developed and validated. Results were disseminated through paired reviewed journals and conferences catching the interest of the international community. Most notable the publication '3D identification of trabecular bone fracture zone using an automatic image registration scheme: a validation study, J Biomechanics. 45 (2012) 2035-2040, related to the validation of the fracture identification procedure. Contacts with several institutions were developed during the project showing particular interest to the identification procedure that showed to be applicable for different kind of problems, even not related to the study of bone structure. Particularly, the micro-CT manufacturer 'Bruker microCT' showed interested in the registration and segmentation procedures and collaboration for the implementation of these tools in the next generation software already started.
The identification procedure developed allowed the study of morphometric characteristic of bone fracture on larger scale. The structural difference of the failure zone, compared to the non-failure was underlined and some particular parameters were identified as dominant for the identification of the weakest point. The project opened the way to the systematic study of trabecular fracture characteristics. This result can, potentially, lead to a new approach of clinical prediction of fracture risk, not based on bone mass, like today, but on the structural parameters highlighted during the study.
Finally, a generalised segmentation mechanism, applicable to both micro and meso-level datasets, was developed and may be used now by other researchers.
The broad objective of the MOSAIC project was to improve the accuracy in the assessment of bone fracture risk by expanding our understanding of bone fracture event. Nowadays, the main investigation technique for the prediction of the fracture event is based on the study of bone quantity. Nonetheless, this approach showed in the past to be unreliable, with errors ranging from 20 to 40 %. The study aimed to develop tools for the identification, study and finally prediction of the trabecular fracture zone. To address these wide objectives, the following scheme was applied:
(a) data collection;
(b) development of image processing techniques;
(c) fracture classification.
Data collection
During the two years of the project, a number of in vitro bone datasets were collected. In the first period of the project 60 datasets were collected from https://www.physiomespace.com/(opens in new window) and produced by Laboratorio Tecnologia Medica, Istituto Ortopedico Rizzoli, Bologna, Italy, with the financial support of the European Union (EU) project 'Living human digital library' (LHDL IST-2004-026932). The datasets were micro-computed-tomography (micro-CT) images of trabecular bone specimen extracted from several different locations from the lower limb of two donors. During the second half of the project, new datasets were collected. The researcher spent four month at the Eindhoven University of Technology (TU), in the Biomechanics and Tissue Engineering (BMTE) section of the department of Biomedical Engineering, under the supervision of Professor Bert Van Rietbergen. BMTE laboratory was involved in the acquisition of the meso-level data within the VPHOP (ICT-2008-223865) EU project. 17 new micro-level datasets of trabecular specimen were collected. Ten of them were available also for meso-level analysis. Finally, 10 datasets of vertebrae were collected both at micro and meso-level.
Development of image processing techniques
The project aimed to the identification, study and prediction of the trabecular fracture. A registration procedure was developed for the identification of the trabecular fracture zone from micro-CT images. The scheme was based on the application of a rigid registration applied on the segmented trabecular structure. The application of the automatic scheme allowed the identification and study of the full three-dimensional (3D) distribution of the structure. The automatic scheme was validated on micro-level datasets. After its validation the scheme was applied on 60 specimens and the morphometric characteristics of trabecular fractures were studied.
Fracture classification
The study and comprehension of trabecular failure zone allowed the development of a fracture classifier. Simple classifiers based on linear regression were investigated to discriminate fractured and non-fractured zone and for the identification of most significant features. The investigation of more sophisticated learning machines allowed applying different classifiers on micro-CT images of a single specimen for the identification of its failure zone. The classifier showed to be able to predict the fracture zone of a single specimen with an accuracy of 98 % over 734 Volume of Interest analysed. Preliminary results were presented at the 5th Workshop on Artificial Neural Networks in Pattern Recognition.
Deviation from the original project
The application of the procedure on meso-level images was not straightforward. Within the project, the proposed fracture identification scheme should be applied on meso-level images. This result was not fully achieved. The reason was that the segmentation process based on fixed thresholding failed the identification of the trabecular structure in the meso-level datasets. Thus, for the half of the second year, the researcher team was concentrated on the development of a new segmentation technique for the meso-level datasets. The technique developed is based on level-set segmentation and initial results were obtained at the end of the project.
Results and potential impact
During the MOSAIC project, a system for the identification, study and prediction of trabecular failure zone in micro-CT images was developed and validated. Results were disseminated through paired reviewed journals and conferences catching the interest of the international community. Most notable the publication '3D identification of trabecular bone fracture zone using an automatic image registration scheme: a validation study, J Biomechanics. 45 (2012) 2035-2040, related to the validation of the fracture identification procedure. Contacts with several institutions were developed during the project showing particular interest to the identification procedure that showed to be applicable for different kind of problems, even not related to the study of bone structure. Particularly, the micro-CT manufacturer 'Bruker microCT' showed interested in the registration and segmentation procedures and collaboration for the implementation of these tools in the next generation software already started.
The identification procedure developed allowed the study of morphometric characteristic of bone fracture on larger scale. The structural difference of the failure zone, compared to the non-failure was underlined and some particular parameters were identified as dominant for the identification of the weakest point. The project opened the way to the systematic study of trabecular fracture characteristics. This result can, potentially, lead to a new approach of clinical prediction of fracture risk, not based on bone mass, like today, but on the structural parameters highlighted during the study.
Finally, a generalised segmentation mechanism, applicable to both micro and meso-level datasets, was developed and may be used now by other researchers.