Decision-Support System for StereoElectroEncephaloGraphy Trajectory Planning

STEREO-ELECTROENCEPHALOGRAPHY (SEEG) is a minimally invasive technique which allows the exploration of brain’s activity in
patients affected by focal epilepsy, helping the identification of the epileptogenic zone (EZ). The procedure requires to implant a variable number of intracerebral electrodes through small holes drilled in the patient’s skull and must accomplish the accurate targeting of the desired intracerebral structures, while minimizing the risk of complications. Traditionally, SEEG planning is performed by a neurosurgeon, who manually selects entry and target points by visually inspecting multi-modal images. Due to the usual number of electrodes (up to 18 per hemisphere) and the need of high accuracy, the planning procedure is complex and very time consuming
(2-3 hours per procedure). Therefore, there is a clear clinical need not covered by commercial planning solutions, that do not provide any advanced assistance nor quantitative information regarding the risk of the planned trajectories.
In this context, the PhD work focuses on the development of a surgical planning decision support system to assist the surgeon during the planning phase of intracerebral electrode trajectories. In particular, the contributions of the thesis are:
1. The identification of the clinical and technological requirements to model the decision-making process in SEEG surgery, providing a generalization of the problem for the analysis of generic image guided percutaneous interventions (C1).
In this phase, we provided the formalization of the planning procedure of percutaneous interventions, as well as we identified the main clinical requirements in terms of safety and efficacy, to be translated into quantitative values with respect to medical image processing and optimization theory. Accordingly, we developed a set of tools which allow the automated definition of optimal trajectories in SEEG by maximizing the distance from vessels, the insertion angle and guarantee anatomy driven explorations. A retrospective quantitative  validation showed that optimized trajectories improved manual planned ones in 98% of the cases in terms of quantitative indexes, even when applying more conservative criteria with respect to actual clinical practice.
2. A novel methodology to exploit retrospective data and manual plans from patient who underwent SEEG procedures in the clinical center (C2), to improve the optimization strategy previously developed. According to the results obtained by C1, we hypothesized that the analysis of retrospective data could support the optimization procedure, providing new information and an objective and reliable method to acquire and transmit the clinical center experience. We developed and analyzed a retrospective database, collecting the data and manually planned trajectories from past cases. Data analysis and machine learning techniques allowed us to obtain the most common paths used by the center to explore specific anatomical brain regions (61 mean trajectories), as well as combinations of  trajectories for macro-areas explorations (8 planning strategies). We developed a specific interface to help the surgeon in the definition of trajectories for new patients, selecting among those obtained by their own data.
3. The development of a decision support system able to exploit the information from past cases and adapt to the specific workflow of the clinical center (C3). The third contribution assembles the development of C1 and C2, since it integrates and extends the two modules. The final system allows the surgeon to define a new patient plan by using the interface developed in C2 and to automatically transfer the trajectories to the target patient anatomy. These initialized trajectories act as spatial and anatomical priors, guiding the optimization procedure initially developed in C1.
The latter has been updated through iterative experiments with our clinical partners, and includes different optimization strategies based
on the trajectory type provided by the initialization module. The validation of the system showed that the initialization module was able
to reproduce 95% of manually planned trajectories, that were subsequently optimized on the patient anatomy. Two surgeons revised 201
optimized trajectories, and considered as clinically feasible 81% of them, while another 7% would have required minor manual adjustments to be used.
The development and results presented in this PhD thesis raised from a strict collaboration between surgeons and engineers, mixing clinical knowledge with medical image analysis, machine learning and optimization theory. The system proposed has been designed to reduce and simplify the interaction between the application and the final user and, therefore, to reduce the planning time while improving precision and safety. The results showed a robust system, able to provide a considerable number of clinically feasible trajectories that only required the surgeon’s revision and approval. It is worth to notice that the implemented system is the result of many iterations
and empirical experiments in collaboration with our clinical partners, that allowed us to better model, implement and identify the combination of clinical requirements and optimization parameters that best fit their workflow. This work set the basis of an adaptable surgical decision support system for SEEG planning, with a modular design and a rich documentation to be continuously improved by the people who has yet to come.