Uploaded January 2026 | Updated September 2026, 2 weeks ago
In this video, Milos Vukadinovic (UCLA) presents 'Vision-language Foundation Models for Cardiac Imaging'.
Presentation Date: 04/02/25
Abstract: With the increasing availability of imaging data and computational power, machine learning has shown promise in automating various aspects of the cardiac imaging pipeline. However, most existing models are task-specific and fail to leverage the full spectrum of available data. There is a need for comprehensive models that synthesize information from all acquired images and report interpretations holistically, much like a cardiologist would. In this talk we will present EchoPrime, a multi-video vision-language model capable of retrieving comprehensive reports for echocardiogram studies. We will discuss what it takes to build a large multimodal model (LMM) and explore its potential integration into clinical practice.
Bio: Milos Vukadinovic is a third-year Bioengineering PhD student at UCLA advised by David Ouyang, a visiting graduate student at Cedars-Sinai, and a data scientist at Kaiser Permanente. He is interested in AI in Medicine applications, and specializes in ML for cardiac medical imaging. Milos is skilled in training large multimodal models, big data analytics and developing ML powered software solutions.
For more information, visit: broadinstitute.org/ml4h
Copyright Broad Institute, 2025. All rights reserved.
In this video, Milos Vukadinovic (UCLA) presents 'Vision-language Foundation Models for Cardiac Imaging'.
Presentation Date: 04/02/25
Abstract: With the increasing availability of imaging data and computational power, machine learning has shown promise in automating various aspects of the cardiac imaging pipeline. However, most existing models are task-specific and fail to leverage the full spectrum of available data. There is a need for comprehensive models that synthesize information from all acquired images and report interpretations holistically, much like a cardiologist would. In this talk we will present EchoPrime, a multi-video vision-language model capable of retrieving comprehensive reports for echocardiogram studies. We will discuss what it takes to build a large multimodal model (LMM) and explore its potential integration into clinical practice.
Bio: Milos Vukadinovic is a third-year Bioengineering PhD student at UCLA advised by David Ouyang, a visiting graduate student at Cedars-Sinai, and a data scientist at Kaiser Permanente. He is interested in AI in Medicine applications, and specializes in ML for cardiac medical imaging. Milos is skilled in training large multimodal models, big data analytics and developing ML powered software solutions.
For more information, visit: broadinstitute.org/ml4h
Copyright Broad Institute, 2025. All rights reserved.
