TractOracle: towards an anatomically-informed reward function for RL-based tractography