Deep learning in real-time ultrasound and digital pathology - From annotation and training to real-world application

Presenter: Erik Smistad, Erik Smistad, Postdoctor. at NTNU CIUS and researcher at SINTEF

Abstract: In this presentation, Erik Smistad will go through some of the many deep learning projects on cardiac ultrasound conducted at the centre of innovative ultrasound solutions (CIUS) over the last six years. This includes image classification, segmentation, optical flow tracking, heart cycle event detection, automatic measurements, guiding and image quality enhancement.  

As ultrasound is a real-time imaging modality we have focused on applying these methods in real-time while the operator is scanning using a our own framework called FAST which can utilize the computational power of modern graphic processing units (GPUs).

The presentation will also cover the recent activities of the AICAN group in Trøndelag which focus on using deep learning methods to analyze digital pathology images. These microscopy images are extremely large and can have a size of up to 200,000 x 100,000 pixels which uncompressed is about 50 gigabytes. This extreme image size makes it hard to store, open and even display these images on a normal computer.

It also poses a large computational challenge when applying deep learning models both in terms of annotation, training and inference. In this presentation, we will demonstrate how we are dealing with these challenges and show our latest software called FastPathology  which enables anyone to apply deep learning models on these large images on normal computers with a simple graphical user interface.

‍

‍

‍

‍

‍

In compliance with GDPRconsent requirements, presentations given in a Visual Intelligence context maybe recorded with the consent of the speaker. All recordings are edited toremove all faces, names and voices of other participants. Questions and commentsby the audience will hence be removed and will not appear in the recording. With the freely given consent from the speaker, recorded presentation may beposted on the Visual Intelligence YouTube channel.

‍