In the last few years, deep learning models have been successfully applied to problems for which we can’t isolate the best imaging biomarkers. Combining rich corpus of dataset and automated data-driven feature extractions algorithms, feature learning using Convolutional Neural Networks (CNN) has shown superiority to hand-crafted and conventional rule-based approaches. By summarizing the data-driven representation of information, that DL model has learned as the best predictors from the raw data, it often reveals unsuspected relations or new trends that lead researchers to better understanding of the complex problems. This departs from the initial use of deep learning in medical imaging as a tool designed exclusively for automation, but it expands the concept to a tool for new discovering of overlooked pattern in the data. In this session examples of data driven imaging feature learning will be presented in the context of medical applications.
Slides: https://ucsf.box.com/s/jpza39x2ntbtai5wo34pya2zv8t0bzzk (MyAccess login required)
The audience should have basic machine learning and medical imaging knowledge.
Dr Pedoia is an Assistant Professor in the Radiology and Biomedical Imaging Department at the University of California San Francisco (UCSF). Her main research interest in developing algorithms for advanced computer vision and machine learning with the aim of improving the usage of non-invasive imaging as diagnostic and prognostic tools. She obtained a PhD in computer science, working on the synoptic description of the 3D information included in fMRI Statistical Parametric Maps through the extraction of salient features and; the development of MRI-based brain segmentation and glial brain tumor classification algorithms. After graduation, she joined the Musculoskeletal and Imaging Research Group at UCSF as post-doctoral fellow. Her role was in providing support and expertise in medical imaging, with a focus to reduce human effort and to extract semantic features from MRI to study degenerative joint diseases. She joined UCSF as a Faculty in July 2018 and her current research focus is on exploring the role of deep learning and multidimensional data analysis in the extraction of contributors to osteoarthritis (OA), by studying analytics to model the complex interactions between morphological, biochemical and biomechanical aspects of the knee joint as a whole; deep learning convolutional neural network for musculoskeletal tissue segmentation and for biomarker discovery.