DGPDT: Detection-Guided and Prompt-Driven Transformers for Automated and Generalizable Cobb Angle Estimation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Accurate Cobb angle measurement is crucial for scoliosis assessment.
• Current methods are labor-intensive and prone to observer variability.

Purpose of the Study:

• To introduce DGPDT (Detection-Guided Prompt-driven Transformer), a unified framework for automated Cobb angle measurement.
• To improve the accuracy and generalizability of scoliosis assessment using AI.

Main Methods:

• Utilized a transformer-based framework integrating vertebra detection (RF-DETR with DINOv2) and foundation-model segmentation (SAM 2.1).
• Employed detection-guided prompting for high-resolution vertebral mask generation.
• Computed Cobb angles from generated vertebral masks for main and compensatory curves.

Main Results:

• Achieved high performance on in-house (TVGH-SpineXR) and external (SpineWeb-16) datasets with Dice coefficients of 0.944 and 0.781.
• Demonstrated mean absolute Cobb angle errors of ~2-3° in-domain and 4.93° cross-domain.
• Showcased accuracy comparable to models trained on benchmark datasets despite limited training data.

Conclusions:

• DGPDT offers a reproducible and clinically acceptable mean absolute error (<5°) for Cobb angle measurement.
• The framework shows potential for broad applicability in spinal analysis beyond the training dataset.
• Automated Cobb angle measurement using DGPDT can enhance scoliosis assessment efficiency and accuracy.