La División Académica de Ingeniería invitan al Seminario "Small Data for Assisted Classification of COVID-19 Patients using X-ray Images" que impartirá el Dr. Edgar F. Román Rangel profesor del Departamento Académico de Computación- ITAM, el proximo viernes 10 de septiembre a las 13:00 hs.

Abstract:

This work introduces a recent collection of chest X-ray images gathered from hospitals in rural areas of Mexico during the Coronavirus 2019 disease (COVID-19) outbreak, and that has been annotated by certified radiologists that classified them into 6 types of respiratory findings related to the COVID-19. This dataset resulted from an altruist effort that brought support to rural areas, enabling them with remote access to expert interpretations of chest X-rays. We present a thorough analysis of several state-of-the-art Convolutional Neural Network (CNN) architectures, which we exploited using transfer learning approaches, and that we adapted to deal with the task of classifying images as positive or negative instances associated with COVID-19. Consistently with one of the main constraints regarding the availability of this type of data worldwide, we also investigated the effects of dealing with limited amounts of data, pushing the training process of these CNN models to the limits of small data by decimating the size of the training set. Our results show that CNN-based methods can achieve classification performance with 0% false negatives rate, while only allowing 6% of overall error, which corresponds to a success rate comparable to human performance.

Bio:

Edgar F. Roman-Rangel is a professor at the Instituto Tecnológico Autónomo de México (ITAM), where he conducts research in computer vision, representation learning, and cross-modal signal processing, applied to cultural and societal challenges. Previous to this appointment, he was a postdoctoral researcher at the University of Geneva from 2013 to 2017, with a break during 2014 and 2015 in which he was a visiting researcher at the National Institute of Anthropology and History of Mexico (INAH). Edgar received his Ph.D. degree in 2013 from the École Polytechnique Fédérale de Lausanne (EPFL) for his work in computer vision and machine learning applied to the analysis of imagery of Mayan hieroglyphs; and he obtained his master diploma in 2006 from the Instituto Tecnológico y de Estudios Superiores de Monterrey (ITESM). Besides his research and academic experience in Computer Vision and Machine Learning, he also has experience also in the banking sector, where he has provided consulting services in statistics and machine learning, and leading development software projects.

Join Zoom Meeting

https://itam.zoom.us/j/99814344763?pwd=ZmJIWklFeDM1NEY0S0x6enk4T09BQT09

Meeting ID: 998 1434 4763

Passcode: 795605