Abstract: Compared with lung cancer, liver cancer and other common cancers, hypopharyngeal cancer is a rare disease. Because the magnetic resonance imaging (MRI) of hypopharyngeal cancer is often uneven, fuzzy and noisy, how to obtain useful information from these MRI images is a difficult problem. It is a major challenge to use deep learning to detect the lesions of hypopharyngeal cancer through MRI images. First, the characteristics and causes of MRI images of hypopharyngeal cancer were summarized, the characteristics and application fields of common target detection networks such as Faster-RCNN, RetinaNet, FCOS, and Cascade-RCNN were then summarized, and the challenges faced by the application of target detection networks in the localization of hypopharyngeal cancer lesions were analyzed. The effective solutions: deformable convolution and application of customized anchors were introduced. Then, the common semantic segmentation networks were introduced, and the challenges of applying these semantic segmentation networks to the segmentation of hypopharyngeal cancer lesions were analyzed. Finally, the target detection network and semantic segmentation network mentioned above were summarized, and the future work of target detection and semantic segmentation of hypopharyngeal cancer medical images was prospected.