Talks and presentations

Learning Representations for Medical Image Diagnosis: Impact of Curriculum Training and Architectural Design

October 14, 2021

Talk, DTIC, Universitat Pompeu Fabra, Barcelona, Spain

This talk summarizes the main outcomes of my PhD thesis. We investigated two key aspects to learn feature representations leveraging Convolutional Neural Networks from medical images for Computer-Aided Diagnosis tasks. In the first part, we explored the role of architectural design in dealing with spatial information. In the second part, we designed curriculum training strategies to control the order, pace, and number of images presented to the optimizer.

Hierarchical deep curriculum learning for the classification of proximal femur fractures

June 24, 2020

Talk, Computer Assisted Radiology and Surgery – CARS, Munich (virtual), Germany

This talk presents a deep curriculum learning strategy that leverages hierarchical information for the fine-grained classification of proximal femur fractures. Our results indicate that using broader labels helps to reduce errors and improve the overall detailed classification.

Precise proximal femur fracture classification for interactive training and surgical planning

April 25, 2020

Talk, International Conference on Information Processing in Computer-Assisted Interventions – IPCAI, Munich (virtual), Germany

This talk presents the feasibility of a fully automatic computer-aided diagnosis tool, based on deep learning, that localizes and classifies proximal femur fractures on X-ray images according to the AO classification. The proposed framework aims to improve patient treatment planning and provide support for the training of trauma surgeon residents. The presentation is available here.

Medical-based Deep Curriculum Learning for Improved Fracture Classification

October 15, 2019

Talk, International Conference on Medical Image Computing and Computer Assisted Interventions – MICCAI, Shenzhen, China

This talk presents strategies derived from knowledge such as medical decision trees and inconsistencies in the annotations of multiple experts, which allow us to assign a degree of difficulty to each training sample. We demonstrate that if we start learning “easy” examples and move towards “hard”, the model can reach a better performance, even with fewer data.

Capsule Networks against Medical Imaging Data Challenges

September 16, 2018

Talk, Large-Scale Annotation of Biomedical Data and Expert Label Synthesis – LABELS Workshop at MICCAI, Granada, Spain

This talk compares the behavior of capsule networks against convolutional neural networks under typical datasets constraints of medical image analysis, namely, small amounts of annotated data and class-imbalance. We evaluate our experiments on MNIST, Fashion-MNIST and medical (histological and retina images) publicly available datasets. Our results suggest that capsule networks can be trained with less amount of data for the same or better performance and are more robust to an imbalanced class distribution, which makes our approach very promising for the medical imaging community.