Detecting and quantifying spatial misalignment between longitudinal kilovoltage computed tomography (kVCT) scans of the head and neck by using convolutional neural networks (CNNs)