Echo from Noise: Synthetic Ultrasound Image Generation Using Diffusion Models for Real Image Segmentation

TY - CHAP

T1 - Echo from Noise

T2 - 4th International Workshop of Advances in Simplifying Medical Ultrasound, ASMUS 2023

AU - Stojanovski, David

AU - Hermida, Uxio

AU - Lamata, Pablo

AU - Beqiri, Arian

AU - Gomez, Alberto

N1 - Publisher Copyright: © 2023, The Author(s).

PY - 2023

Y1 - 2023

N2 - We propose a novel pipeline for the generation of synthetic ultrasound images via Denoising Diffusion Probabilistic Models (DDPMs) guided by cardiac semantic label maps. We show that these synthetic images can serve as a viable substitute for real data in the training of deep-learning models for ultrasound image analysis tasks such as cardiac segmentation. To demonstrate the effectiveness of this approach, we generated synthetic 2D echocardiograms and trained a neural network for segmenting the left ventricle and left atrium. The performance of the network trained on exclusively synthetic images was evaluated on an unseen dataset of real images and yielded mean Dice scores of 88.6 ± 4.91, 91.9 ± 4.22, 85.2 ± 4.83 % for left ventricular endocardium, epicardium and left atrial segmentation respectively. This represents a relative increase of 9.2, 3.3 and 13.9% in Dice scores compared to the previous state-of-the-art. The proposed pipeline has potential for application to a wide range of other tasks across various medical imaging modalities.

AB - We propose a novel pipeline for the generation of synthetic ultrasound images via Denoising Diffusion Probabilistic Models (DDPMs) guided by cardiac semantic label maps. We show that these synthetic images can serve as a viable substitute for real data in the training of deep-learning models for ultrasound image analysis tasks such as cardiac segmentation. To demonstrate the effectiveness of this approach, we generated synthetic 2D echocardiograms and trained a neural network for segmenting the left ventricle and left atrium. The performance of the network trained on exclusively synthetic images was evaluated on an unseen dataset of real images and yielded mean Dice scores of 88.6 ± 4.91, 91.9 ± 4.22, 85.2 ± 4.83 % for left ventricular endocardium, epicardium and left atrial segmentation respectively. This represents a relative increase of 9.2, 3.3 and 13.9% in Dice scores compared to the previous state-of-the-art. The proposed pipeline has potential for application to a wide range of other tasks across various medical imaging modalities.

KW - Diffusion Models

KW - Image synthesis

KW - Ultrasound

UR - http://www.scopus.com/inward/record.url?scp=85174716747&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-44521-7_4

DO - 10.1007/978-3-031-44521-7_4

M3 - Conference paper

AN - SCOPUS:85174716747

SN - 9783031445200

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 34

EP - 43

BT - Simplifying Medical Ultrasound - 4th International Workshop, ASMUS 2023, Held in Conjunction with MICCAI 2023, Proceedings

A2 - Kainz, Bernhard

A2 - Müller, Johanna Paula

A2 - Kainz, Bernhard

A2 - Noble, Alison

A2 - Schnabel, Julia

A2 - Khanal, Bishesh

A2 - Day, Thomas

PB - Springer Science and Business Media Deutschland GmbH

Y2 - 8 October 2023 through 8 October 2023

ER -