Robust CNN multi-nested-LSTM framework with compound loss for patch-based multi-push ultrasound shear wave imaging and segmentation | JoVE Visualize

Area of Science:

Medical Imaging
Biomedical Engineering
Artificial Intelligence in Medicine

Background:

Ultrasound shear wave elastography (SWE) offers noninvasive tissue elasticity assessment but faces challenges in reconstruction accuracy, noise sensitivity, and data limitations.
Current SWE methods struggle with scalable region coverage and reliable segmentation of inclusions, impacting diagnostic reliability.
Limited annotated data hinders the generalization and robustness of existing SWE reconstruction algorithms.

Purpose of the Study:

To develop a novel two-stage deep learning framework for enhanced SWE reconstruction and inclusion segmentation.
To address limitations of noise sensitivity, inefficient multi-push strategies, and data scarcity in SWE.
To improve the accuracy and reliability of quantitative elasticity measurements for tissue pathology assessment.

Main Methods:

A two-stage deep learning framework combining a CNN-based multi-nested-LSTM reconstruction network and a compound-loss-driven CNN-denoiser.
The reconstruction stage utilizes a ResNet3D-encoder, Nested CNN-LSTM, temporal attention, and frequency attention for feature extraction and elasticity mapping.
The second stage employs a dual-decoder denoising network for inclusion and background stiffness, with a fusion module for denoised maps and segmentation masks, trained with a patch-based regime.

Main Results:

The framework achieved high performance on simulated and experimental phantom data, with peak-signal-to-noise-ratio (PSNR) up to 26.33 dB and intersection over union (IoU) up to 0.813.
Experimental validation on swine liver demonstrated accurate elasticity estimates within biological ranges.
The approach outperformed existing methods like DSWE-Net and spatio-temporal CNNs in reconstruction, segmentation, and noise insensitivity.

Conclusions:

The proposed deep learning framework offers a robust solution for SWE reconstruction and inclusion segmentation.
The method shows significant potential for improving quantitative elasticity measurements and clinical translation in medical diagnostics.
This work advances the field of ultrasound elastography by providing a more accurate and reliable tool for tissue pathology assessment.