Timely and accurate diagnosis in medical imaging for head and neck cancer (HNC) is important for better patient management and regime development. This work provides a critical assessment of deep learning techniques for diagnosing HNC in computed tomography (CT) and magnetic resonance imaging (MRI) images. For HNC detection, a DL-based prototype system is built dedicated to HNC using state-of-the-art architectures such as U-Net, R2U-Net, Attention U-net, and Attention R2U-Net. Extensive validation is carried out on an extensive database that comprises CT/MRI series of patients with HNC. By comparing the performance of these different DL architectures, the current study determines efficient models of HNC detection and the impact of the imaging methods, namely CT and MRI scans, on the performance of the models. The discussion presented outlines the possibilities and initiatives for the implementation of the investigated DL-based approaches in improving HNC diagnosis and underpins the creation of effective CAD tools for clinical practice. Attention R2U-Net achieved the best accuracy of 99.48% in recognizing MRI scans when separation split the data into 80% for training and 20% for testing. All MRI-based implementation models achieved slightly better results than CT-based implementation models during every assessment attempt.