IEEE IEEE 14th IEEE EMBS-SPS International Summer School on Biomedical Imaging
Saint-Jacut de la Mer, Emerald Coast, Brittany
France, 19-25 June, 2022
Claudia Mazo Vargas

Claudia Mazo Vargas

PhD. Claudia Mazo is a Research Fellow at University College Dublin (UCD) since 2021. From 2017 to 2021, she was a Marie Skłodowska-Curie Postdoctoral Fellow with UCD and Oncomark Ltd, Ireland. From 2017 to 2018, she worked as a researcher at Vicomtech in San Sebastian, Spain, within the eHealth and Biomedical Applications Area. She received a double-PhD, the first an Engineering in Production and Computing Doctorate from the Universidad de León, Spain, in 2016, and the second an Engineering Doctorate with Emphasis in Computer Science from the Universidad del Valle, Colombia, in 2017. From 2009 to 2011, she worked as a young researcher at the Universidad del Valle with the Multimedia and Computer Vision Group.

She has been the primary author on numerous publications in national and international conferences and high-impact journals. Her research interests include machine learning, image processing, pattern recognition, computer vision, and personalised medicine.

Claudia's awards and honours include:


Machine Learning for Histopathological Images

Claudia Mazo Vargas
Research Fellow, University College Dublin., Dublin - Ireland
claudia.mazovargas@ucd.ie

Abstract

Some computer vision problems, such as those operating on histopathological images, consist of identifying distinctive patterns and structures of interest to infer a tissue/organ or to analyse a biomarker. These have been challenging for researchers and still remain open problems. Since an organ is a 3D structure and an image is a 2D representation of an organ sample, inferring information about these organs from their representation is an inverse problem. Considering also that information about the physiological functions of these organs is not directly contained in the image, these problems are often ill-posed and their solutions unstable.

To further complicate matters, current real-world applications need interdisciplinary teams (i.e., containing pathologists, statisticians, morphologists, entrepreneurs, neurosurgeries, psychologists, bioscientists, bioinformatics, neurologists, among others) which arise from different backgrounds and generate heterogeneous types or representations of data that cannot operate or be treated in the same way. Heterogeneous data can be used to improve the performance by achieving lower margins of error or uncertainties than those solutions using a single information source. These kinds of proposals constitute a new way to integrate multiple heterogeneous information sources and to unify them in a manner that is imperceptible to users but still beneficial to them. For instance, in the context of breast cancer, this information could include: (i) pathological data such as tumour size and grade, ER, HER2, and Ki67 status; (ii) standard clinical data; (iii) environmental data; (iv) demographic and lifestyle risk factors (sex, age, weight, race/ethnicity, alcohol consumption, smoking, exercise, family, and personal history); and (v) histopathological knowledge representations, all of which need to be interpreted and processing by a machine.

Additionally, I want to share my own experience to motivate and guide postgraduate students to continue working on biomedical image analysis and get to know about acclaimed research fellowships, such as the Marie Sklodowska-Curie Postdoctoral Fellowship.


version 4.6, jun., 2022