Segmentation with Speckle Reduction and Superresolution by Deep Leaning for Human Ultrasonic Echo Image

Area of Science:

• Medical imaging
• Artificial intelligence in medicine
• Biomedical engineering

Background:

• Ultrasound (US) imaging is crucial for diagnosing human malignant tissues, relying on morphological interpretation of echo images.
• Image speckles in ultrasound hinder accurate tissue segmentation, posing a challenge for diagnosis.
• Deep learning (DL) offers a novel approach to image denoising, surpassing traditional signal processing methods.

Purpose of the Study:

• To investigate the application of deep learning (DL) for reducing speckles in ultrasound images.
• To enhance the segmentation accuracy of the carotid artery lumen using DL techniques.
• To evaluate the effectiveness of DL-based superresolution for improving echo image quality.

Main Methods:

• Utilized DL for denoising ultrasound images to reduce speckles.
• Applied DL segmentation algorithms, commonly used for medical images, to the denoised ultrasound data.
• Implemented DL superresolution to further improve segmentation accuracy, particularly for echo images.
• Validated the proposed methods through simulations and in vivo experiments.

Main Results:

• DL denoising successfully reduced speckles in ultrasound images.
• DL segmentation achieved improved accuracy on denoised images.
• DL superresolution further enhanced segmentation performance for the carotid artery lumen.
• In vivo experiments confirmed the effectiveness of the developed methods.

Conclusions:

• DL-based denoising, segmentation, and superresolution offer a powerful toolkit for improving ultrasound image analysis.
• The proposed approach enhances the accuracy of carotid artery lumen segmentation.
• Method effectiveness is confirmed for in vivo data, demonstrating clinical relevance.