Duke Center for Health Informatics: Attention Based Multiple Instance Learning and its Application in Imaging Informatics

October 5, 2022
4:00 pm to 5:00 pm
Webcast

Event sponsored by:

School of Medicine (SOM)
Duke Center for Health Informatics

Contact:

Johnstone, Jessica

Share

Speaker:

Khalid Niazi, PhD

Oct 5 Recording

Seminar Abstract: Dr. Khalid Niazi will discuss the fundamentals of attention-based multiple instance learning (AB-MIL) followed by its application to histopathology, anesthesiology, radiology, and sleep medicine. Unlike strictly supervised classification, which assumes labels related to the object dominate the overall image, the AB-MIL paradigm allows for heterogeneous objects' existence in one image. AB-MIL treats the whole image as a “bag” with multiple objects called “instances” from different implicit classes. I will demonstrate how this model can be used to discover novel imaging biomarkers for disease classification. The presentation will also exhibit how it can be modified to infer gene expression values from histopathology images. Finally, the presentation will show how this model can be used to reduce intra-reader variability.

Instructor Biosketch: Dr. Niazi is an Assistant Professor of Biomedical Informatics and Internal Medicine at Wake Forest School of Medicine. He is responsible for developing image analysis methods (aka deep learning methods), algorithms, and systems for biomedicine. His unique educational training and professional experience enables him to conduct collaborative and meaningful research with academic institutions, industry, and the business sector. Over the years, he has devised novel digital image analysis methods that enable computers to solve intriguing clinical and research problems that directly or indirectly impact clinical care. Unlike conventional methods, these methods are inspired by neurophysiology, sociology, nature and physics. His research focuses on enabling the computers to learn complex concepts and tasks from extraordinarily small empirical samples. He has been developing data management and analysis technologies, to address problems with the integration of data from different modalities. This is a critical step needed to address some of the pressing biomedical research questions today. Dr. Niazi’s contributions with core fundamentals of medicine and the statistical underpinnings of data analytics and knowledge discovery are regularly published in prestigious journals and conferences, and his research is funded by NIH, DoD, and several foundation grants.


Informatics Research Seminars