Predictive uncertainty: Knowing what we don’t know in image data

Predictive uncertainty: Knowing what we don’t know in image data

Sponsored by

Digital image analysis using AI has great utility to standardize data, increase biomarker discovery and aid in diagnostic assessment of disease. For image analysis in clinical applications, AI development relies on indication-specific training sets to develop and train the algorithm, as well as separate images for testing. However, once the algorithm leaves the development team, there is an unavoidable element of uncertainty from the vastly greater pool of image sets that algorithm will encounter.

• How do we know we can trust the models used to develop the APP?
• Is there a way to test the confidence predictions coming from these models?

In this talk, we will provide insights into how we work with concepts of uncertainty and how these affect our development of APPs and datasets for assisting primary diagnosis. We will also give examples from our recent research, development and validation of our lymph node metastasis detection AI APP.

Date: 15 December 2020
Time: New York – 11:00 | London – 16:00 | Paris – 17:00 | Singapore – 00:00 (16 December) |Tokyo – 01: 00 (16 December) |Sydney – 03:00 (16 December)
Duration: 60 minutes
Event structure: 30-minute presentation + 30-minute Q&A
Registration fee: Complimentary access
Webinar on-demand: Available to view until midnight 29 December 2020 (registration is required)