Recognizing Focal Liver Lesions in Contrast-Enhanced Ultrasound with Discriminatively Trained Spatio-Temporal Model

TL;DR

This paper introduces a discriminative spatio-temporal model for automatic detection of focal liver lesions in contrast-enhanced ultrasound videos, improving diagnostic assistance with a novel inference algorithm.

Contribution

It presents a new framework modeling lesion regions as latent variables and combines spatial-temporal pruning for efficient inference, advancing automated FLL diagnosis.

Findings

Achieves promising results on the largest publicly available dataset

Effectively models spatial and temporal enhancement patterns

Improves diagnostic accuracy for FLLs in CEUS videos

Abstract

The aim of this study is to provide an automatic computational framework to assist clinicians in diagnosing Focal Liver Lesions (FLLs) in Contrast-Enhancement Ultrasound (CEUS). We represent FLLs in a CEUS video clip as an ensemble of Region-of-Interests (ROIs), whose locations are modeled as latent variables in a discriminative model. Different types of FLLs are characterized by both spatial and temporal enhancement patterns of the ROIs. The model is learned by iteratively inferring the optimal ROI locations and optimizing the model parameters. To efficiently search the optimal spatial and temporal locations of the ROIs, we propose a data-driven inference algorithm by combining effective spatial and temporal pruning. The experiments show that our method achieves promising results on the largest dataset in the literature (to the best of our knowledge), which we have made publicly available.