Tweet
24 months - Post-Doctoral Fellowship
Bone fracture reduction in trauma:
A combined Deep-Learning/Statistical Shape Model approach for the estimation of the optimal reduction and fixation strategy during planning
Scientific context
Bone trauma is a global scourge and ranks as the third leading cause of overall morbidity burden according to the World Health Organization (WHO). Contrary to the scheduled orthopedic surgery that benefits from recent efficient edge-cutting digital technologies, trauma constraints are higher with a shorter period of management, a high complexity and multiplicity of treated cases and a greater variability of the surgeon experience. To this end, a funded research project “ReBone” (ANR RHU, 24M€) gathering industrial, academic, and medical partners have been recently launched. The aim is to optimize the surgery and minimize complications in complex bone trauma by developing and validating personalized, automated, and collaborative pre-operative planning solutions, and by providing intraoperative solutions allowing the accurate execution of the planning. “ReBone” will provide thus a fundamental, deep, and continuous improvement in clinical and surgical management of severe trauma and complex fractures, to (1) reduce treatment delays, (2) have access to more precise and reproducible guided surgeries, (3) decrease post-operative complications. The Laboratory of Medical Information Processing (LaTIM) will be in charge of the fracture reduction modelling and the simulation of the hardware fixation (plate and screws). Still today, most of the available commercial solutions to plan bone fracture reduction rely only on manual approaches which is hardly acceptable for the surgeon especially when the number of fractures to be treated becomes too large. Our aim is therefore to provide a complete easy-to-use planning software solution allowing the automatic and fast simulation of both the fracture reduction and the hardware fixation based on geometrical criteria.
Mission
Most of the existing methods for the simulation of the fracture reduction are based on the incorporation of the mirrored healthy contralateral bone. But these methods cannot be automatically used in all patients, especially in cases of natural shape differences or bilateral trauma [6]. Others approaches based on the usual registration algorithms have been also proposed to estimate the reduction process from the alignment of the fracture lines, but their performances were very dependent on the quality of the initialization. Statistical Shape Models (SSMs) have been recently introduced. Although SSMs do not require contralateral bones and can better drive the estimation of the fracture alignment, the accuracy of the estimation is very dependent on the class and type of the fractures. We want therefore to propose a generic approach allowing the accurate, robust, and fast simulation of both the optimal fracture reduction and hardware fixation whatever the type and class of the fracture. Combined approaches exploiting both Deep-Learning and shape priors will be investigated to overcome current limitations. Validation will be carried out by comparing the proposed results with manual reductions and fixation strategies proposed by surgeons on real data. An integration of the proposed approach will be finally integrated, in collaboration with engineers, inside our planning software to be used by surgeons.
Environment
This post-doc position will be hosted in the LaTIM. Born from the complementarity between health and data science, the LaTIM laboratory develops multi-disciplinary research driven by members from IMT Atlantique, CHRU Brest, University of Western Brittany and Inserm. The recruited postdoc will work in collaboration with academic, industrial and hospital partners within the context of the RHU Rebone project. Access will be given to clinical data from our clinical partners as well as to the PLaTIMed platform (https://platimed.fr/) to make realistic evaluation of the proposed approaches.
Profile
PhD in computer vision, AI, applied mathematics.
Good programming skills is an important requisite, especially in python and C++. Autonomy, open-mindedness, and motivation
Good English skills are also expected.
Application
CV with list of publications, cover letter and two letters of recommendation, have to be sent to Guillaume Dardenne (guillaume.dardenne@inserm.fr)
Bone fracture reduction in trauma:
A combined Deep-Learning/Statistical Shape Model approach for the estimation of the optimal reduction and fixation strategy during planning
Scientific context
Bone trauma is a global scourge and ranks as the third leading cause of overall morbidity burden according to the World Health Organization (WHO). Contrary to the scheduled orthopedic surgery that benefits from recent efficient edge-cutting digital technologies, trauma constraints are higher with a shorter period of management, a high complexity and multiplicity of treated cases and a greater variability of the surgeon experience. To this end, a funded research project “ReBone” (ANR RHU, 24M€) gathering industrial, academic, and medical partners have been recently launched. The aim is to optimize the surgery and minimize complications in complex bone trauma by developing and validating personalized, automated, and collaborative pre-operative planning solutions, and by providing intraoperative solutions allowing the accurate execution of the planning. “ReBone” will provide thus a fundamental, deep, and continuous improvement in clinical and surgical management of severe trauma and complex fractures, to (1) reduce treatment delays, (2) have access to more precise and reproducible guided surgeries, (3) decrease post-operative complications. The Laboratory of Medical Information Processing (LaTIM) will be in charge of the fracture reduction modelling and the simulation of the hardware fixation (plate and screws). Still today, most of the available commercial solutions to plan bone fracture reduction rely only on manual approaches which is hardly acceptable for the surgeon especially when the number of fractures to be treated becomes too large. Our aim is therefore to provide a complete easy-to-use planning software solution allowing the automatic and fast simulation of both the fracture reduction and the hardware fixation based on geometrical criteria.
Mission
Most of the existing methods for the simulation of the fracture reduction are based on the incorporation of the mirrored healthy contralateral bone. But these methods cannot be automatically used in all patients, especially in cases of natural shape differences or bilateral trauma [6]. Others approaches based on the usual registration algorithms have been also proposed to estimate the reduction process from the alignment of the fracture lines, but their performances were very dependent on the quality of the initialization. Statistical Shape Models (SSMs) have been recently introduced. Although SSMs do not require contralateral bones and can better drive the estimation of the fracture alignment, the accuracy of the estimation is very dependent on the class and type of the fractures. We want therefore to propose a generic approach allowing the accurate, robust, and fast simulation of both the optimal fracture reduction and hardware fixation whatever the type and class of the fracture. Combined approaches exploiting both Deep-Learning and shape priors will be investigated to overcome current limitations. Validation will be carried out by comparing the proposed results with manual reductions and fixation strategies proposed by surgeons on real data. An integration of the proposed approach will be finally integrated, in collaboration with engineers, inside our planning software to be used by surgeons.
Environment
This post-doc position will be hosted in the LaTIM. Born from the complementarity between health and data science, the LaTIM laboratory develops multi-disciplinary research driven by members from IMT Atlantique, CHRU Brest, University of Western Brittany and Inserm. The recruited postdoc will work in collaboration with academic, industrial and hospital partners within the context of the RHU Rebone project. Access will be given to clinical data from our clinical partners as well as to the PLaTIMed platform (https://platimed.fr/) to make realistic evaluation of the proposed approaches.
Profile
PhD in computer vision, AI, applied mathematics.
Good programming skills is an important requisite, especially in python and C++. Autonomy, open-mindedness, and motivation
Good English skills are also expected.
Application
CV with list of publications, cover letter and two letters of recommendation, have to be sent to Guillaume Dardenne (guillaume.dardenne@inserm.fr)