Uncertainty Quantification in Medical Image Analysis

MICCAI 2024 Tutorial

2nd edition

10 October 2024

8 AM - 12 PM

Day 2 of Satellite events

About

This MICCAI 2024 half-day event aims to foster discussions within the scientific community regarding the trustworthiness of deep learning model predictions by providing valuable insights into uncertainty quantification and model calibration techniques necessary for making reliable decisions.

This second edition of the tutorial will be focused on the topic of uncertainty quantification, model calibration, and conformal prediction. This year's tutorial is conducted in a joint event with the UNSURE workshop, combining talks from experts, hands-on materials, modern research presentation, keynote presentation, and a panel discussion.

Participants will have an opportunity to gain a comprehensive understanding of uncertainty quantification, covering mathematical foundations and cutting-edge research questions while acquiring practical tools for deriving state-of-the-art uncertainty estimates in medical image analysis.

We believe this tutorial is an essential resource for scientists, clinicians, and industry professionals aiming to enhance the robustness and reliability of their DL models.

Learning objectives

This tutorial will enable participants to gain knowledge about topics ranging from the very fundamentals of uncertainty quantification to cutting-edge recent methods in the field. We expect tutorial attendees to develop:

• A solid understanding of the basic concepts around uncertainty quantification in machine learning and of a taxonomical map of the field.

• The ability to discriminate between different sources of uncertainty.

• Mathematical fundamentals for understanding and implementing several well-established uncertainty quantification techniques

• Understanding of the UQ methods used for the medical image analysis tasks

• Knowledge of the tasks of robustness to and detection of Out-of-Distribution data and Model Calibration.

• Knowledge of uncertainty quantification through the basics of Conformal Prediction.

Program

8:00-8:30 Part 1: General Introduction into Uncertainty & Domain Shift by Vatsal Raina

8:30-9:00 Part 2: Uncertainty for Medical Image Analysis by Nataliia Molchanova

9:00-9:15 Part 3: Lightning Session: Model Calibration by Meritxell Riera i Marín

9:15-9:30 Part 4: Lightning Session: Conformal Prediction by Adrian Galdran

9:30-10:00 Spotlight presentation session by UNSURE Workshop participants

10:00-10:30 Coffee break with a poster session

10:30-11:05 Keynote session: The multi-dimensional aspect of uncertainty in automated medical image analysis by Michel Dojat

11:05-12:05 Oral presentation session by UNSURE Workshop participants

12:05-12:30 Panel discussion and closing session

Keynote presentation

Michel Dojat

Research Director at Inserm and Deputy Scientific Director for digital biology and health at Inria

His major scientific fields are Neuroimaging, Neuroinformatics, Image and Signal analysis with a specific focus on Human Vision. He is involved in different AI in medicine projects and has developed specific tools for brain imaging analysis and ICU monitoring. He is the scientific manager of FLI-IAM, a national project for the management of in vivo imaging data, metadata and processing pipelines. He contributes to specific actions for supporting data sharing for clinical and preclinical studies. He is cofounder and scientific advisor of Pixyl, an AI-based medical services company.

He is IEEE senior fellow. He received his engineer diploma in Material Physics from Insa (Lyon, Fr, 1982), a PhD in Computer Science (Paris, Fr, 1994) and a Habilitation à Diriger des Recherches (Grenoble, Fr, 1999).

List of publications: GoogleScholar.

Minimal required skills

This tutorial does not require any prior skills in uncertainty or model calibration, however some background in deep learning may be needed.

Materials

All the materials of the previous year (UQ Tutorial MICCAI 2023) and this year are available in our GitHub. There you can find presentations and Jupyter Notebooks with hands-on. Several video recordings are already available on YouTube. References on UQ can be found in the UQ in MIA Cookbook. More materials will be provided closer to the tutorial date.