Synthetic Boost: Leveraging Synthetic Data for Enhanced Vision-Language Segmentation in Echocardiography

TL;DR

This paper demonstrates that pretraining vision-language segmentation models on synthetic echocardiography images generated by diffusion models improves segmentation accuracy and convergence speed, addressing data scarcity issues.

Contribution

It introduces a novel approach of using synthetic data from Semantic Diffusion Models to enhance vision-language segmentation in echocardiography.

Findings

Improved segmentation metrics with synthetic pretraining

Faster convergence during model training

Effective use of language prompts for model guidance

Abstract

Accurate segmentation is essential for echocardiography-based assessment of cardiovascular diseases (CVDs). However, the variability among sonographers and the inherent challenges of ultrasound images hinder precise segmentation. By leveraging the joint representation of image and text modalities, Vision-Language Segmentation Models (VLSMs) can incorporate rich contextual information, potentially aiding in accurate and explainable segmentation. However, the lack of readily available data in echocardiography hampers the training of VLSMs. In this study, we explore using synthetic datasets from Semantic Diffusion Models (SDMs) to enhance VLSMs for echocardiography segmentation. We evaluate results for two popular VLSMs (CLIPSeg and CRIS) using seven different kinds of language prompts derived from several attributes, automatically extracted from echocardiography images, segmentation masks, and their metadata. Our results show improved metrics and faster convergence when pretraining VLSMs on SDM-generated synthetic images before finetuning on real images. The code, configs, and prompts are available at https://github.com/naamiinepal/synthetic-boost.