Vertebra-Level Completeness Analysis in Thoracolumbar Ultrasound Using a YOLO-Based Detection Framework

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Ultrasound offers radiation-free scoliosis monitoring but faces challenges with rib shadowing and speckle noise obscuring vertebrae.
• Existing deep learning methods for scoliosis assessment focus on localization accuracy, neglecting anatomical completeness.
• Accurate assessment of vertebral structures is crucial for effective scoliosis management.

Purpose of the Study:

• To develop and evaluate a deep learning model for assessing vertebral completeness in ultrasound scans for scoliosis.
• To introduce the Vertebra Presence Matrix (VPM) for explicit quantification of anatomical completeness.
• To improve automated scoliosis evaluation by integrating completeness metrics with localization accuracy.

Main Methods:

• A YOLO-based detection framework was combined with the Vertebra Presence Matrix (VPM) for a vertebra-level completeness model.
• Thoracolumbar ultrasound scans were annotated and split into training/testing sets using a patient-wise approach.
• Four model variants were evaluated, including full-spine and vertebra-centric crop representations with single-class and 17-class detection heads.

Main Results:

• The 17-class vertebra-centric crop model demonstrated superior stability compared to the full-spine model, especially in areas of high anatomical variability.
• The 17-class crop model achieved a mean Average Precision at 50% IoU (mAP50) of 0.929.
• A scan-level completeness score of 0.74 was attained using the Vertebra Presence Matrix (VPM).

Conclusions:

• Vertebral completeness in ultrasound-based scoliosis monitoring can be explicitly quantified using the developed model and VPM.
• Integrating completeness metrics with localization accuracy enhances automated scoliosis evaluation.
• The proposed approach offers a more comprehensive assessment of anatomical structures in ultrasound imaging for scoliosis.