Domain Adaptation in Multi-Modal Data

 Personnel: Yash-yee Logan, Mohit Prabhushankar

Goal: To characterize possible domain differences on OLIVES Dataset

Challenges: Domain difference in ophthalmic data can arise from sources such as treatment, disease manifestation, and clinical labels. 

Our Work: We study the domain adaptation between the first and last week’s worth of data in terms of DR/DME detection [1]. Training and testing within the first week data provides the best results for biomarker detection. We adapt between the first and last visit domains. Specifically, we use a part of the last visit data to train with the first visit data and show that: a) adapting between OCT scans before and after treatment is possible, and of DR/DME.

References:

[1] M. Prabhushankar, K. Kokilepersaud*, Y. Logan*, S. Trejo Corona*, G. AlRegib, C. Wykoff, "OLIVES Dataset: Ophthalmic Labels for Investigating Visual Eye Semantics," in Advances in Neural Information Processing Systems (NeurIPS 2022) Track on Datasets and Benchmarks, New Orleans, LA, Nov. 29 - Dec. 1 2022. [PDF]