Joint Spine Segmentation and Noise Removal From Ultrasound Volume Projection Images With Selective Feature Sharing.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Spine Diagnostics

Background:

• Volume Projection Imaging (VPI) from ultrasound data is crucial for visualizing spine features and diagnosing Adolescent Idiopathic Scoliosis (AIS).
• Current methods may face challenges with scan noise and precise feature segmentation, impacting diagnostic accuracy.
• Intelligent scoliosis assessment requires robust imaging techniques that can overcome these limitations.

Purpose of the Study:

• To develop a novel multi-task deep learning framework for simultaneous noise reduction and spine feature segmentation in VPI.
• To improve the accuracy and efficiency of Adolescent Idiopathic Scoliosis diagnosis using ultrasound VPI.
• To provide an alternative intelligent solution for clinical scoliosis assessment.

Main Methods:

• A dual-stream framework incorporating a weakly-supervised generative adversarial network for noise removal.
• A spine segmentation stream to accurately predict bone masks.
• A selective feature-sharing strategy to enable effective interaction between noise removal and segmentation tasks.

Main Results:

• The proposed framework demonstrated promising performance in both scan noise removal and spine feature segmentation.
• Weakly-supervised noise reduction was achieved without requiring paired noisy-clean image samples.
• Accurate segmentation of spine features was achieved, facilitating intelligent scoliosis assessment.

Conclusions:

• The novel multi-task framework offers an effective approach for enhancing ultrasound VPI for scoliosis diagnosis.
• The method successfully addresses scan noise and improves spine feature segmentation.
• This technique presents a valuable tool for facilitating clinical diagnosis of Adolescent Idiopathic Scoliosis.