Evaluating accuracy in artificial intelligence-powered serial segmentation for sectional images applied to...

Area of Science:

• Medical Imaging
• Computational Anatomy
• Artificial Intelligence in Medicine

Background:

• Three-dimensional (3D) reconstruction is crucial for anatomical and histological studies but is often hindered by time-consuming manual segmentation.
• Artificial intelligence (AI) offers potential solutions to automate and improve the efficiency of segmentation processes.

Purpose of the Study:

• To evaluate the accuracy of the AI-based segmentation and tracking model SAM-Track for anatomical and histological structures.
• To explore the potential of AI in enhancing research efficiency for 3D reconstruction.

Main Methods:

• SAM-Track was used to segment anatomical and histological structures from computed tomography (CT), magnetic resonance imaging (MRI), and cadaveric sections.
• Segmentation accuracy was quantified using the Dice similarity coefficient, comparing AI-generated masks with manual segmentations.
• Segmented images were reconstructed in 3D to assess shape fidelity.

Main Results:

• SAM-Track demonstrated variable accuracy (Dice scores 0.13-0.83) for CT and MRI, with higher scores for clear-edged structures (e.g., femur, liver).
• Histological sections showed high accuracy (e.g., tibia 0.95, pancreas 0.90), particularly for well-delineated tissues.
• Tracking of branching structures like arteries and veins had lower success rates (0.72, 0.52).

Conclusions:

• AI-based automatic segmentation using SAM-Track provides moderate-to-good accuracy for diverse anatomical and histological structures.
• The model shows promise for improving efficiency in morphological studies requiring 3D reconstruction.
• Further development may be needed for optimal performance on soft tissues and complex vascular networks.