Evaluating the impact of reinforcement learning on automatic deep brain stimulation planning.

PURPOSE: Traditional techniques for automating the planning of brain electrode placement based on multi-objective optimization involving many parameters are subject to limitations, especially in terms of sensitivity to local optima, and tend to be replaced by machine learning approaches. This paper explores the feasibility of using deep reinforcement learning (DRL) in this context, starting with the single-electrode use-case of deep brain stimulation (DBS). METHODS: We propose a DRL approach based on deep Q-learning where the states represent the electrode trajectory and associated information, and actions are the possible motions. Deep neural networks allow to navigate the complex state space derived from MRI data. The chosen reward function emphasizes safety and accuracy in reaching the target structure. The results were compared with a reference (segmented electrode) and a conventional technique. RESULTS: The DRL approach excelled in navigating the complex anatomy, consistently providing safer and more precise electrode placements than the reference. Compared to conventional techniques, it showed an improvement in accuracy of 2.3% in average proximity to obstacles and 19.4% in average orientation angle. Expectedly, computation times rose significantly, from 2 to 18 min. CONCLUSION: Our investigation into DRL for DBS electrode trajectory planning has showcased its promising potential. Despite only delivering modest accuracy gains compared to traditional methods in the single-electrode case, its relevance for problems with high-dimensional state and action spaces and its resilience against local optima highlight its promising role for complex scenarios. This preliminary study constitutes a first step toward the more challenging problem of multiple-electrodes planning.

Generation of synthetic training data for SEEG electrodes segmentation.

PURPOSE: Stereoelectroencephalography (SEEG) is a minimally invasive surgical procedure, used to locate epileptogenic zones. An accurate identification of the metallic contacts recording the SEEG signal is crucial to ensure effectiveness of the upcoming treatment. However, due to the presence of metal, post-operative CT scans contain strong streak artefacts that interfere with deep learning segmentation algorithms and require a lot of training data to distinguish from actual contacts. We propose a method to generate synthetic data and use them to train a neural network to precisely locate SEEG electrode contacts. METHODS: Random electrodes were generated following manufacturer's specifications and dimensions and placed in acceptable regions inside metal-free CT images. Metal artefacts were simulated in the generated data set using radon transform, beam hardening, and filtered back projection. A UNet neural network was trained for the contacts segmentation task using various training set-ups combining real data, basic augmented data, and synthetic data. The results were compared. RESULTS: We reported a higher accuracy when including synthetic data during the network training, while training only on real and basic augmented data more often led to misclassified artefacts or missed contacts. The network segments post-operative CT slices in less than 2 s using 4 GeForce RTX2080 Ti GPUs and in under a minute using a standard PC with GeForce GTX1060. CONCLUSION: Using synthetic data to train the network significantly improves contact detection and segmentation accuracy.