Spotting L3 slice in CT scans using deep convolutional network and transfer learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate anatomical landmark identification is crucial for quantitative analysis in CT scans.
• Manual slice localization is time-consuming and prone to inter-observer variability.
• Automated methods can enhance efficiency and consistency in radiological workflows.

Purpose of the Study:

• To develop and validate a fully automated system for precise slice localization in 3D CT scans.
• To identify the third lumbar vertebra (L3) slice, a representative anatomical marker.
• To create a system adaptable to various patient body regions without prior assumptions.

Main Methods:

• A machine learning regression approach utilizing transfer learning from pre-trained deep architectures (ImageNet).
• A three-step pipeline: CT to Maximum Intensity Projection (MIP) conversion, Convolutional Neural Network (CNN) prediction via sliding window, and sequence analysis for slice height determination.
• Application to L3 slice detection in 642 patient CT scans.

Main Results:

• Achieved an average localization error of 1.91±2.69 slices (less than 5 mm).
• Processing time averaged less than 2.5 seconds per CT scan.
• Demonstrated high accuracy and speed suitable for clinical integration.

Conclusions:

• The automated system reliably and efficiently detects specific anatomical slices, such as the L3 vertebra, in CT scans.
• The use of transfer learning significantly reduces the need for extensive annotated training data.
• The proposed method offers a robust and rapid solution for slice localization, facilitating clinical applications.