Hyperbolic Vision Language Models (VLMs) in medical imaging

Hyperbolic Vision Language Models (VLMs) in medical imaging

Type: Master CS

Period: TBD

Student: (Unassigned)

If you are interested please contact :

dr. A. Briasouli PhD (Alexia) Assistant Professor

+31534896985

 a.briassouli@utwente.nl

Building: Zilverling 4049

Personal page

Hyperbolic VLMs have recently been shown to surpass Euclidean VLMs by providing richer semantic representations¹. A novel hyperbolic density embedding based image-text representation learning approach tailored for specific medical domain data HYDEN: Hyperbolic Density Representations for Medical Images and Reports2 has also been proposed. The resulting embeddings are shown to provide better discrimination between classes and improve classification and retrieval performance. In this project, you should examine hyperbolic vision-language embeddings across different distribution-shift scenarios in medical data.

REFERENCES:

1. Intriguing Properties of Hyperbolic Embeddings in Vision-Language Models
2. HYDEN: Hyperbolic Density Representations for Medical Images and Reports
3. Delving into Out-of-Distribution Detection with Medical Vision-Language Models
4. Interpreting Biomedical VLMs on High-Imbalance Out-of-Distributions: An Insight into BiomedCLIP on Radiology
5. Generalized Out-of-Distribution Detection and Beyond in Vision Language Model Era: A Survey
6. OpenMIBOOD: Open Medical Imaging Benchmarks for Out-Of-Distribution Detection
7. Medical Out-of-Distribution Analysis Challenge