Dental diseases, such as caries and periodontal diseases, are among the most common non-communicable diseases for human beings. Radiographic imaging is a necessary tool used by dental clinicians to view underlying dental structures, for diagnosis and treatment planning, with millions of dental radiographs being produced in the United Kingdom each year. Surface level digital imaging is also used to document dental health and model surface level structures. However, there are well documented and significant inconsistencies among dentists in assessing the same radiograph and average detection specificity rates are low due to various issues, including the subtlety of diseases, quality of radiographs, data heterogeneity, and varying expertise of individual dentists. Existing deep learning approaches focus primarily on the direct detection of diseases of similar features with high inter class similarity. However, contextual information and feature rich anatomies are usually not considered, despite contributing to the fine localisation and staging of the disease in clinical practice. Most structural anatomies and diseases also often follow a clear object hierarchy that can be introduced into detection methodologies. Applications of deep learning for this domain often disregards the methodology used by clinicians to make a diagnosis. This work aims to develop novel structure aware and hierarchy aware deep learning methodologies, that focuses on providing clinically relevant information in a manner that encourages human computer collaboration and streamlines clinical processes. Identification of diseases through dental structures could enable increased detection and staging performance, automation of time consuming processes and increase model explainability via the design of the detection method itself. This work also explores the development of hierarchy aware methodologies, that leverage the inherent dependencies of coarse parent classes to detect fine grained child classes, and the injection of heuristic domain knowledge during model training or during inference.