Backdoor Mitigation in Deep Neural Networks via Strategic Retraining