**7.1. Conclusion**

We proposed an approach to accurate and real-time simulation of coil deployment, as well as prediction of coil embolization outcome in a patient-specific environment for clinical planning and rehearsal.


#### **7.2. Perspectives**

Regarding future work, first of all, we are further improving the computational efficiency, since real-time simulation has not been achieved when performed on a mesh consisting of over 8,000 elements. We still want to more deeply investigate various computational strategies to obtain real-time computation by using more advanced numerical schemes, such as parallel implementation on GPU of the backtracking step, optimization of linear solvers by the preconditioning technique coupled with a GPU-based conjugate gradient implementation [3]. Additionally, more advanced techniques to generate a multi-resolution mesh, such as adaptive refinement [32], could also be a solution, as it maximizes the result accuracy while minimizing the computational effort.

Of course, we acknowledge that further validation is required, both on the DEC method and on other elements of the simulation. However, comparison between simulated results and *in vivo* data remains difficult due to the limitations of current medical imaging techniques. We are starting to investigate the use of fluoroscopic imaging to assess the simulation of medical devices used in interventional radiology procedures. Regarding the specific validation of the flow computation, a possible direction is MR imaging which, under some modalities, can measure flow patterns. Although quite noisy, such data could provide interesting insights and help validate our results.

Before put to wide clinical utilization, our method should be further improved. The boundary conditions obtained from patient-specific real data will improve the realism of the simulated results. To obtain such real data, we can benefit from the imaging techniques or the state-of-the-art medical instruments, such as the ultra-miniature pressure catheter, which is a so small sensor on the catheter that it does not significantly change the blood flow. Furthermore, standards of assessment and metrics of evaluation should be set up to offer doctor the information on the minimal number of coils required to achieve embolization in long term, and to evaluate the doctor's performance on the procedure.

Finally, when using stent-assisted coiling, the stent acts as scaffolding, and prevents the coils from falling out. Our work could be easily extended to simulate the outcome by taking the collision contact between coils and the stent into account.
