top of page
[X] logo.jpg

Masha Baidachna

Computing Science

Year of study:


Mirror, Mirror on the Wall: Automating Dental Smile Analysis with AI in Smart Mirrors


Driven by the demand for prompt diagnosis, this study integrates artificial intelligence into dental aesthetics, with a focus on smile analysis. The aim is to confront prevalent challenges such as latency, subjectivity, and the lack of smart diagnostic tools in dental aesthetics. Enabled by the advancements in generative artificial intelligence (GenAI), neural networks, and the Internet of Things, we propose a framework to detect smile aesthetic issues such as excessive gingival display (i.e. a high lip line that exposes 2-3 mm of gingiva above the crown of the maxillary teeth) and smile arc classification. Despite recent improvements in deep learning within computer vision, the dental sector contains a gap of publicly available datasets. To mitigate overfitting risks, our approach incorporates GenAI data augmentation to improve convolutional neural network (CNN) performance. We leverage a diffusion model to generate a text-to-image dataset utilized across training, validation, and testing phases in different proportions. Notably, the GenAI CNN model exhibits an impressive accuracy of 81.607% in detecting excessive gingival display, surpassing established baselines. The development of a fully connected pipeline from the CNN model to the user interface facilitates timely interactions between patients and clinics. This framework unveils the potential of a visionary Internet of Mirrors, offering a precise diagnostic tool and fostering enhanced communication between users and healthcare providers.


I am a second year computer science student from Ukraine. I finished high school in the US and moved to Glasgow for university. I have always been fascinated by mathematics, particularly linear algebra, so machine learning naturally appealed to me. I started my research as an EPSRC summer intern with the School of Engineering and gained a deeper understanding and appreciation for the power of neural networks. I continued the research, aiming towards publication, and delved into becoming a reviewer for an ICLR workshop on large language models. I intend to keep exploring and contributing to this dynamic field in the future.

bottom of page