OpenArm 2.0: Automated Segmentation of 3D Tissue Structures for Multi-Subject Study of Muscle Deformation Dynamics

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Artificial Intelligence

Background:

• Accurate segmentation of musculoskeletal structures is crucial for biomechanical analysis.
• Existing methods for 3D ultrasound segmentation often require extensive manual annotation.
• Developing efficient and accurate AI models for musculoskeletal imaging is an ongoing challenge.

Purpose of the Study:

• To develop a novel neural-network-based pipeline for segmenting 3D muscle and bone structures from 2D ultrasound data of the human arm.
• To optimize the pipeline's performance through data augmentation and strategic training data selection.
• To create and release the OpenArm 2.0 dataset for research in muscle force modeling and assistive device control.

Main Methods:

• Utilized the U-Net neural network framework for image segmentation.
• Investigated various data augmentation techniques to enhance model performance.
• Developed strategies for efficient training data selection to minimize manual annotation.
• Generated the OpenArm 2.0 dataset with multi-subject, multi-angle, and multi-force scans.

Main Results:

• A novel neural-network pipeline for 3D ultrasound segmentation of arm structures was successfully developed.
• Optimized performance was achieved through careful selection of data augmentation and training datasets.
• The OpenArm 2.0 dataset, featuring volumetric annotations of biceps and humerus, was created.
• The dataset is publicly available to facilitate further research.

Conclusions:

• The developed AI pipeline offers an effective solution for segmenting musculoskeletal structures from ultrasound data.
• The OpenArm 2.0 dataset represents a significant resource for advancing muscle force modeling and related fields.
• This work paves the way for improved assistive device control and biomechanical simulations.