From 2D Ultrasound to Patient-Specific 3D Surface Models for Interventional Guidance.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Deep Learning in Medicine

Background:

• Traditional liver ultrasound lacks 3D anatomical context, hindering follow-up studies and interventional accuracy.
• Current methods require CT or MR scans for liver volume measurements, increasing patient burden.
• No existing solutions provide real-time 3D anatomical context for large organ ultrasound scanning.

Purpose of the Study:

• To develop a novel approach for constructing patient-specific 3D liver models using B-mode ultrasound and positional tracking.
• To provide 3D anatomical context during ultrasound examinations for improved accuracy and patient comfort.
• To enable precise lesion localization for interventional procedures like biopsies.

Main Methods:

• Utilizing state-of-the-art deep learning algorithms for liver surface landmark detection.
• Employing positional sensor tracking data alongside B-mode ultrasound images.
• Registering a geometric model to a surface point cloud to generate patient-specific 3D liver models.
• Implementing a semi-automated workflow for model creation and validation.

Main Results:

• Successful generation of patient-specific 3D anatomical surface models of the liver from ultrasound data.
• Demonstrated potential for improved accuracy in interventional guidance and lesion localization.
• Achieved model accuracy within acceptable limits for clinical application.

Conclusions:

• The proposed method offers a novel solution for creating 3D liver models from ultrasound, addressing limitations of current techniques.
• This 3D anatomical context is crucial for enhancing ultrasound-guided interventions and patient care.
• The semi-automated workflow ensures reliable and accurate patient-specific 3D liver models for clinical use.