Dr Giacomo Tarroni

Overview

Giacomo Tarroni has been a full-time researcher in the field of medical image analysis since 2009. His work has been mainly focused on image segmentation, image registration, quality control and object tracking for cardiovascular, brain and fetal images. In particular, he obtained his Ph.D. from the University of Bologna, Italy (in collaboration with the University of Chicago, U.S.) working on the automated analysis of first-pass myocardial perfusion sequences in MRI. During his post-doc at the University of Padova, Italy, he focused on the automated analysis of fetal ultrasound images. After being awarded a Marie Skłodowska-Curie Fellowship from the European Commission, he moved to Imperial College London, where he became interested in the applications of machine learning and AI to automated organ detection, quality control assessment and motion correction for cardiac MRI.

His current research focus is on machine learning approaches for unsupervised anomaly detection, self-supervised image classification/segmentation and federated learning, both for medical image analysis and more generally for computer vision applications.

Giacomo was able to generate over 183 k€ in funding from the EU by winning a Marie Skłodowska-Curie Fellowship for the project JUNO, of which he was Principal Investigator. In addition, in his career he has collaborated with several high-profile medical image analysis research projects, including SmartHeart (EPSRC, 5M£), iFind (Wellcome Trust and EPSRC , 10M£), CHIRON (EU Artemis-JU, 18M€).

Giacomo has recently been Associate Editor for the IEEE ISBI 2019 conference, and he often acts as reviewer for international journals (e.g. IEEE Trans Med Imaging, Medical Image Analysis, IEEE Trans Image Process, Plos ONE) as well as renowned international conferences (e.g. MICCAI, IEEE ISBI). Since 2018 he is also board member of the UK Chapter of Marie Curie Alumni Association (MCAA-UK).

Giacomo is a member of the CitAI Research Centre.

Qualifications

1. PhD, University of Bologna, Italy, Jan 2009 – Jun 2012
2. M.S. in Electronic Engineering, University of Bologna, Italy, Sep 2005 – Oct 2008
3. B.S. in Electronic Engineering, University of Bologna, Italy, Sep 2002 – Aug 2005

Employment

1. Lecturer in Artificial Intelligence, City, University of London, Sep 2019 – present
2. Research Fellow, Imperial College London, Nov 2017 – present
3. Marie Skłodowska-Curie Fellow, Imperial College London, Nov 2015 – Nov 2017
4. Post-doctoral Research Associate, University of Padova, Jun 2013 – Nov 2015
5. Post-doctoral Research Associate, University of Bologna, Jul 2012 – May 2013
6. Visiting Researcher, University of Chicago, Apr – Jul 2010

Unsupervised anomaly detection for medical images

Modern machine learning approaches (i.e. deep learning methods) have been recently proposed to automatically identify anomalies in
medical images. Most of them are based on supervised
learning and consequently have two important constraints.
First, they require large and diverse annotated datasets for
training. Second, they are specific to the abnormalities
annotated in the training datasets and therefore are unable to
generalise to other pathologies. Unsupervised anomaly
detection methods aim to overcome these
constraints by not relying on annotated datasets. Instead,
they focus on learning the underlying distribution of normal
images and then identifying as anomalies the images that do
not conform to the learnt distribution.

Our research is aimed at identifying novel methods for unsupervised anomaly detection in medical images, e.g. using vector-quantised variational auto-encoders and implicit field learning. We have recently published these approaches on top conferences in medical image analysis (e.g. ISBI and MICCAI).

Self-supervised medical image analysis

In computer vision tasks, self-supervision enables neural networks to leverage available unlabelled datasets for pre-training. Many different approaches recently proposed in the literature (e.g. SimCLR) are capable of reaching the accuracy of fully-supervised counterparts for image classification and segmentation.

Our research is aimed at developing novel approaches for self-supervised image classification/segmentation in the context of medical images.

Federated learning for medical image processing

In most applications, machine learning is performed in a centralised way: data is pulled from the different sources and model training happens in a single node. This setting is however unrealistic in most modern scenarios, where 1) data owners don’t want/can’t share their data and 2) data distribution varies from source to source. Federated learning consists in strategies aiming at addressing these issues. The core idea is that data remain at the different sources and only model updates are shared.

Our research is aimed at developing novel strategies for federated learning in the scenario of medical image classification/segmentation.

Featured publications

1. Marimont, S.N. and Tarroni, G. (2021). Anomaly Detection Through Latent Space Restoration Using Vector Quantized Variational Autoencoders. 2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI) 13-16 April.
2. Tarroni, G., Oktay, O., Sinclair, M., Bai, W., Schuh, A., Suzuki, H. … Rueckert, D. (2018). A Comprehensive Approach for Learning-Based Fully-Automated Inter-slice Motion Correction for Short-Axis Cine Cardiac MR Image Stacks.
3. Bai, W., Suzuki, H., Huang, J., Francis, C., Wang, S., Tarroni, G. … Rueckert, D. (2020). A population-based phenome-wide association study of cardiac and aortic structure and function. Nature Medicine, 26(10), pp. 1654–1662. doi:10.1038/s41591-020-1009-y.
4. Biffi, C., Cerrolaza, J.J., Tarroni, G., Bai, W., Marvao, A.D., Oktay, O. … Rueckert, D. (2020). Explainable Anatomical Shape Analysis through Deep Hierarchical
   Generative Models. IEEE Transactions on Medical Imaging. doi:10.1109/TMI.2020.2964499.
5. Tarroni, G., Oktay, O., Bai, W., Schuh, A., Suzuki, H., Passerat-Palmbach, J. … Rueckert, D. (2019). Learning-Based Quality Control for Cardiac MR Images. IEEE Transactions on Medical Imaging, 38(5), pp. 1127–1138. doi:10.1109/tmi.2018.2878509.
6. Tarroni, G., Corsi, C., Antkowiak, P.F., Veronesi, F., Kramer, C.M., Epstein, F.H. … Patel, A.R. (2012). Myocardial Perfusion: Near-automated Evaluation from Contrast-enhanced MR Images Obtained at Rest and during Vasodilator Stress. Radiology, 265(2), pp. 576–583. doi:10.1148/radiol.12112475.

Publications by category