Tutorials

Two tutorial sessions are programmed for Sunday, 8th September, 2024.


Tutorial 1

Dr. Adrien Depeursinge

Institute of Informatics, HES-SO Valais/CHUV, Techo-Pole 3, 3960 Sierre, Switzerland.

Dr. Mario Jreige

Department of Nuclear Medicine and Molecular Imaging, CHUV, Rue du Bugnon 46, 1011 Lausanne, Switzerland

Dr. Vincent Andrearczyk

Institute of Informatics, HES-SO Valais, Techo-Pole 3, 3960 Sierre, Switzerland.

Radiomics: success stories, negative results, challenges ahead and hands-on sessions

Time

9:00 – 12:30.

Abstract

Almost fifteen years after the first papers coined the word “radiomics”, the first part of this tutorial will review the most notable achievements to date, FDA and CE approved products addressing image-based personalized medicine, clinical adoption, as well as notable negative results and areas where initial promises have not been fulfilled. We will then summarize best practices, study quality assessment and identify paradigms that are most appropriate for building successful radiomics models. We will then present our experience as the main organisers of the HEad and neCK TumOR segmentation and outcome prediction (HECKTOR) challenge for three consecutive years, our coordination efforts in the context of the second chapter of the Image Biomarker Standardisation Initiative (IBSI), our experience in the design and use of anthropomorphic phantoms to assess the reproducibility and harmonisation of radiomics. We will conclude this first part with perspectives in the context of approaches able to aggregate longitudinal patterns from multiple lesions for assessing prognosis and response to treatment. In the second part, we will organise hands-on sessions using our QuantImage v2 web platform, where participants will have the opportunity to develop, validate and interpret radiomics models, starting from pre-extracted features from a dataset of more than 100 patients. We will conclude the tutorial with a participatory discussion of the hands-on experience and general concluding remarks.

Speakers’ Bios

Prof. Adrien Depeursinge is a full professor at the Institute of Informatics of the HES-SO Valais-Vallis, Switzerland since 2014 and a Group leader at the Department of Nuclear Medicine and Molecular Imaging, CHUV since 2018. He has more than 85 journal publications in the domain of artificial intelligence for image-based personalized medicine and organized several events on the topic in international conferences and networks.

Dr. Mario Jreige is an attending physician and senior lecturer specialized in radiology and nuclear medicine at the Department of Nuclear Medicine and Molecular Imaging of the Lausanne University Hospital (CHUV). With a keen interest in radiomics research, he has published several articles on artificial intelligence in medical imaging. He has participated in the organization of several international conferences on this topic and is currently leading multiple projects in the field.

Dr. Vincent Andrearczyk received a double Masters degree in electronics and signal processing from INP-ENSEEIHT, France and Dublin City University, in 2012 and 2013 respectively. He completed his PhD degree on deep learning for texture and dynamic texture analysis at Dublin City University in 2017. He is currently a senior researcher at the University of Applied Sciences and Arts Western Switzerland with a research focus on medical imaging and model interpretability.

Tutorial 2

Dr. Reza Farahani

Institut für Informationstechnologie, Alpen-Adria-Universität Klagenfurt

Dr. Vignesh V Menon

Video Communication and Applications Department, Fraunhofer HHI, Berlin, Germany

A Tutorial on Latency- and Energy-Aware Video Coding and Delivery Streaming Systems

Time

9:00 – 12:30.

Abstract

This tutorial introduces modern performance and energy-aware video coding and content delivery solutions and tools, focusing on popular video streaming applications, i.e., VoD and live streaming. In this regard, after introducing fundamentals of modern video encoding and networking paradigms, we introduce modern solutions systems, using per-title encoding, per-scene encoding, virtualized and software networks, edge computing, overlay networks such as Content Delivery Networks (CDNs) and/or Peer-to-Peer (P2P) paradigms to provide latency and energy-efficient VoD and live HAS streaming. Furthermore, the tutorial also presents our tools, software, datasets, and testbeds to demonstrate our latest achievements and share practical insights for researchers, engineers, and students who want to improve conversational streaming or even test such techniques for immersive video sequences (e.g., tile-based 360-degree VR) with a focus on latency, economic cost, and energy.

Speakers’ Bios

Dr. Reza Farahaniis a Postdoctoral researcher at the Institute of Information Technology (ITEC), University of Klagenfurt. He received his Ph.D. and M.Sc degrees in information technology and computer engineering in 2023 and 2019, respectively, from the University of Klagenfurt, Klagenfurt, Austria, and the University of Tehran, Tehran,IRAN. From 2019 to 2023, he was involved in the ATHENA project, funded by the Christian Doppler Forschungsgesellschaft and the industry partner Bitmovin GmbH. From November 2022 to January 2023, he was a visiting scholar at the 5G&6G Innovation Centre (5GIC&6GIC), Institute for Communication Systems (ICS), at the University of Surrey, UK. He has co-authored over 25 publications in international conferences and journals. Additionally, he has held the role of technical program chair for several international conferences. He has also worked in the computer networks field in different roles, e.g., network engineer, developer, protocol designer, and instructor (R&S, SP) for over six years. His research interests are Network and Service Management, Parallel and Distributed Systems, Multimedia Communication, Edge-Cloud Continuum, Serverless Computing, Network Softwarization and Virtualization, Mathematics Optimization, and Distributed Learning approaches.

Dr. Vignesh V Menon is a Postdoctoral researcher at the Video Communication and Applications Department, Fraunhofer HHI, Berlin. He received his Ph.D. from the University of Klagenfurt, Austria. He received a B.Tech. in Electronics and Communication Engineering from Amrita Vishwa Vidyapeetham University, India, and an M.Sc. in Information and Network Engineering from KTH Royal Institute of Technology, Sweden, in 2016 and 2020, respectively. He worked as a Project Assistant at Christian Doppler Laboratory ATHENA from 2020 to 2023. He also worked as a Software Engineer developing video encoding software solutions in MulticoreWare Inc., India, between 2016-2018 and Divideon, Sweden, between 2018-2020. He has co-authored four patents and over 30 publications in esteemed conferences and journals. He received the best paper award at the ACM Mile High Video Conference 2024. His research interests are image and video compression and video streaming.


Tutorial 3

Dr. Abolfazl Mehranian

General Electric HealthCare (GEHC), UK

Deep learning-assisted quantitative image reconstruction in SPECT and PET

Time

13:30 – 17:00.

Abstract

In this tutorial, we first review the life cycle of AI/ML solutions (from data collection to model training and deployment) in order to establish a foundation for reviewing different categories of deep learning (DL)-based methods for PET and SPECT image reconstruction and enhancement. We will review state-of-the-art solutions and DL architectures and finally will touch upon commercially available AI solutions for PET and SPECT.

Dr Mehranian is the lead AI scientist for molecular imaging at GE healthCare. He is an expert in medical image reconstruction and responsible for designing, training and deployment of AI solutions at GEHC. He received his PhD in medical physics from University of Geneva in 2015 under supervision of Prof Habib Zaidi. He then joined Prof Andrew Reader’s lab at King’s College London to continue his research in PET-MR image reconstruction. He has received a number of prestigious awards including GEHC Global MICT recognition award in 2023, IEEE Young Investigator Award in 2019 and The Researcher of the Year in 2018 at School of Biomedical Engineering and Imaging Sciences, King’s College London.

Tutorial 4

Meritxell Bach Cuadra

CIBM Center for Biomedical Imaging, Lausanne University (UNIL), Radiology Department (CHUV)

Pedro M. Gordaliza

CIBM Center for Biomedical Imaging, Lausanne University (UNIL), Radiology Department (CHUV)

Jonathan Rafael-Patiño

EPFL

Nataliia Molchanova

CIBM Center for Biomedical Imaging, Lausanne University (UNIL), Radiology Department (CHUV)

Vladyslav Zalevskyi

CIBM Center for Biomedical Imaging, Lausanne University (UNIL), Radiology Department (CHUV)

Margaux Roulet

CIBM Center for Biomedical Imaging, Lausanne University (UNIL), Radiology Department (CHUV)

Distributions Shifts, Generalizability and Reproducibility in Brain MRI Analysis

Time

13:30 – 17:00.

Abstract

The application of computer vision and deep learning to brain MRI analysis is hindered by key challenges:
1. Small Dataset Sizes and Data Scarcity: Brain imaging datasets are very small, often containing only hundreds or thousands of samples, compared to large natural image datasets. This data scarcity limits the training of accurate, generalizable models. Data sharing across studies is critical but remains limited due to privacy constraints and lack of standardization.
2. Data Bias and Distribution Shifts: MRI acquisition protocols vary significantly, leading to systematic biases and distribution shifts in the data. Models trained on one distribution fail to generalize well to others.
3. Lack of Reproducibility: Due to the challenges above and several image acquisition artefacts such as subject movement, hardware failures, technician experience, etc.

This tutorial will cover cutting-edge techniques to address these challenges, including:
– Efficient image quality control and artifact detection
– Uncertainty quantification for model predictions
– Causal reasoning to understand bias sources and debiasing techniques
– Data augmentation using synthetic data and simulating distribution shifts
– Quantitative MRI harmonization across scanners and protocols
– Federated learning for collaborative model training while preserving privacy

Speaker’s Bio

Meritxell Bach Cuadra

Pedro M. Gordaliza obtained his BSc in Telecommunication engineering from Universidad Carlos III de Madrid (UC3M). After graduating in 2010, he worked for four years as a data scientist in the telecommunication industry in Spain and the UK. In 2014 he started a new position as a research engineer in the UC3M and Gregorio Marañón hospital lab (Laboratorio de Imagen Médica, LIM) to develop image processing algorithms for disease prediction and quantification for several medical imaging modalities (i.e., MRI, US, PET-CT). During this period, in 2018, he obtained his MSc in Advanced Artificial Intelligence. In 2022 Pedro completed his PhD in Biomedical Engineering, defending the PhD thesis entitled: “Computer-Aided Assessment of Tuberculosis with Radiological Imaging: From rule-based methods to Deep Learning“.  Pedro joined the CIBM SP CHUV-UNIL team headed by Meritxell Bach Cuadra as a postdoctoral researcher in October 2022. His research focuses on Domain Adaptation for brain imaging analysis and its links with uncertainty prediction and causal representation learning.

Jonathan Rafael-Patiño is a Postdoctoral Researcher at the École Polytechnique Fédérale de Lausanne (EPFL). His research focuses on diffusion MRI, Federated Learning, and MRI Simulations. He is particularly interested in the application of advanced machine learning techniques and convex optimization to study the brain’s microstructure and its impact on cognitive behavior. His recent publications include studies on the heterogeneity of diffusion imaging protocols and their impact on clinical assessments in acute stroke, as well as developing computational frameworks for realistic white matter microstructure substrates.

Nataliia Molchanova received a bachelor degree in Physics with a minor in mathematical modelling at Moscow State University in 2019. That year she started a master program in Computational science and engineering at Ecole Polytechnique Fédérale de Lausanne (EPFL). Since March 2022 Nataliia is working on a PhD thesis funded by Haslerfoundation within the project “Explaining AI decisions in personalized healthcare: towards integration of deep learning into diagnosis and treatment planning (MSxplain)”. Her PhD is focused on developing uncertainty and explainability strategies for AI methods to existing deep learning models for MR image analysis in Multiple Sclerosis diagnosis, such as white matter and cortical lesion detection and paramagnetic rim classification.

Vladyslav Zalevskyi graduated in 2017 from Igor Sikorsky Kyiv Polytechnic Institute in Ukraine where he received a bachelors degree in Computer Science with a focus on applied systems in artificial intelligence. Afterwards, he pursued an Erasmus Mundus Joint Masters degree in Medical Imaging and Applications (MAIA) between University of Burgundy (France), University of Girona (Spain) and University of Cassino (Italy). He wrote his master thesis at the Danish Research Center for Magnetic Resonance (DRCMR) in Copenhagen, which focused on the use of synthetic data and contrastive self-supervised learning. Since,  September 2023, Vladyslav joined a project that focuses on exploring and tackling domain shifts in fetal brain imaging under supervision of Dr. Meritxell Bach Cuadra, funded by the Swiss National Science Foundation, exploring further how different approaches like synthetic image generation and self-supervised learning can be used to make deep learning models robust to domain shifts present in pediatric MRI.

Margaux Roulet holds a MSc. Degree in Life Science Engineering from the École Polytechnique Fédérale de Lausanne. During her studies she specialized in Biomedical Engineering which provided her with a solid foundation in biomedical imaging, neurotechnology, and AI-related topics. Margaux’ academic and industrial experiences has fostered her keen interest for translational clinical research. She carried out her master thesis at NeuroRestore, CHUV, where she developed a closed-loop algorithm for a rehabilitation task for patients with spinal cord injuries. She further joined Neurosoft Bioelectronics as an R&D Engineer to develop a flexible subdural electrode technology and took part in pre-clinical research projects in collaboration with EPFL. At the moment, Margaux is pursuing her PhD thesis focusing on tackling domain shifts in the reconstruction, segmentation and quantitative mappings between newborn and fetal populations and between high (1.5T & 3T) and low (0.55T) field strengths. The thesis is under the supervision of Dr. Meritxell Bach Cuadra with the support of Swiss National Science Foundation.