Enhanced Deep-Learning-Based Automatic Left-Femur Segmentation Scheme with Attribute Augmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate segmentation of the left femur in computed tomography (CT) scans is crucial for orthopedic analysis and surgical planning.
• Current deep learning models for femur segmentation can be limited by variations in patient positioning and data quality.

Purpose of the Study:

• To enhance the performance of a deep-learning-based automatic left-femur segmentation scheme.
• To investigate the impact of augmenting CT slices with data attributes, specifically lying position, on segmentation accuracy.

Main Methods:

• A deep learning model was developed for automatic left-femur segmentation.
• CT datasets were categorized into eight groups (F-I-F-VIII) based on data augmentation strategies.
• Segmentation performance was evaluated using Dice Similarity Coefficient (DSC), Intersection over Union (IoU), Spectral Angle Mapper (SAM), and Structural Similarity Index Measure (SSIM).

Main Results:

• The highest segmentation performance, measured by DSC (88.25%) and IoU (80.85%), was achieved using cropped and augmented CT datasets with large feature coefficients (Category F-IV).
• The model demonstrated strong similarity between predicted and ground-truth 3D reconstructions, with SAM values between 0.117-0.215 and SSIM values between 0.701-0.732.
• Attribute augmentation in medical image preprocessing proved effective in improving segmentation outcomes.

Conclusions:

• Augmenting CT slices with data attributes, such as lying position, is a novel and effective strategy for enhancing deep learning-based automatic left-femur segmentation.
• This approach leads to more accurate segmentation and reliable 3D reconstructions, with potential benefits for clinical applications.
• The study highlights the importance of data preprocessing techniques in optimizing deep learning model performance for medical image analysis.