31 datascEYEnce! Computer Vision News Last but not least, Nacho mentioned the development of a multi-modal foundation model. With the help of LLMs, the retinar team uses a concept of crawling through the internet and getting all sorts of retinal images with accompanied text to characterize the images. In the next step, the foundation model will be trained in a self-supervised setup and later on fine-tuned with a restricted dataset to better adapt to the target clinical scenarios. After hearing so much about various data processing on the platform, an obviously important topic is patient privacy. Nacho explained that the private clinical report stays private until shared by the patient with a specialist. Furthermore, a patient can give consent for each image to be used for research purposes, enabling future investigations and improvements. As a final remark, I want to mention quite a nice phrase from Nacho that we should all keep in mind - we are working on data of human beings with feelings, family, friends and life goals. Being able to increase their quality of life and making an impact on treatment outcomes is a great motivation for him and hopefully for all of us! Mercedes Leguía, leading ophthalmologist of retinar, acquiring fundus pictures for the project at Hospital El Cruce (Argentina).
RkJQdWJsaXNoZXIy NTc3NzU=