The Statistical Atlases and Computational Modeling of the Heart (STACOM) workshop has been running annually at MICCAI since 2010. The 10th edition of STACOM workshop will be held on 13 October 2019 at the MICCAI 2019 in Shenzhen, China. The STACOM workshop is aiming to create a collaborative forum for young/senior researchers (engineers, biophysicists, mathematicians) and clinicians, working on: statistical analysis of cardiac morphology and dynamics, computational modelling of the heart and fluid dynamics, data/models sharing, personalisation of cardiac electro-mechanical models, quantitative image analysis and translational methods into clinical practice.
Disentangled representation learning in medical imaging
Professor Sotirios A. Tsaftaris (University of Edinburgh, UK)
The detection of disease, segmentation of anatomy and other classical analysis tasks, have seen incredible improvements due to deep learning. Yet these advances need lots of data: for every new task, new modality, new hospital more training data are needed. In this talk, I will show how deep neural networks can learn latent and disentangled embeddings suitable for several analysis tasks in the heart. Within a multi-task learning setting I will show that the same framework can learn embeddings drawing supervision from self-supervised tasks that use reconstruction and also temporal dynamics overall reducing considerably need for annotation. I will discuss how different architectural choices can solve key problems in multimodal data processing and critically also allow us to learn reducing need for image-level annotations by obtaining supervision also from text-based health records. I will conclude by highlighting challenges for deep learning in healthcare in general.
We provide 45 multi-sequence CMR images from patients who underwent cardiomyopathy. Each patient had been scanned using the three CMR sequences, i.e. the LGE, T2 and bSSFP. The task of this challenge is to segment the ventricles and myocardium from LGE CMR, combing with other two sequences (T2 and bSSFP) from same patients, which can be used to assist the LGE CMR segmentation.
Go to challenge website
The aim of this challenge is to learn effective machine learning models that can estimate a set of clinical significant LV indices (regional wall thicknesses, cavity dimensions, area of cavity and myocardium, cardiac phase) directly from MR images. No intermediate segmentation is required in the whole procedure.
The main goal of the challenge is to assess the ability of cardiac computational models to predict the electrical outcome of pacemaker-based therapies. The challenge will be based on multi-modal data acquired at Hospital Clínic de Barcelona of a swine model of left bundle branch block for experimental studies of CRT. Biventricular finite-element meshes, including cardiomyocyte orientation from rule-based models and local activation times mapped from electro-anatomical data at LBBB and CRT stages, will be provided to the participants as training data (4 cases). After model personalization, participants will be asked to predict electrical response in the testing dataset (8 cases).
The STACOM 2019 workshop accepts regular paper submission describing new methods in the following (not limited) topics:
- Statistical analysis of cardiac morphology and morphodynamics
- Computational modeling and simulation of the heart and the great vessels
- Personalisation of cardiac model, electrophysiology and mechanics
- Quantitative cardiac image analysis
- Sharing and reusing cardiac model repository
- Translational studies of cardiac image analysis in clinical practice