Automatic lower limb bone segmentation in radiographs with different orientations and fields of view based on a...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Automating bone identification and segmentation in X-ray images is vital for orthopedic clinical procedures.
• Current methods often require manual intervention, limiting efficiency.

Purpose of the Study:

• To develop an automated method for identifying and segmenting lower limb bone structures in radiographs.
• To improve the accuracy and efficiency of bone segmentation in diverse X-ray imaging conditions.

Main Methods:

• A modified SegNet neural network with a wide contextual architecture was employed for pixel-wise semantic segmentation.
• The network captures both global and local contextual information for enhanced accuracy.
• Additional labels were used to manage overlapping bone structures.

Main Results:

• The approach achieved an average detection rate of 98.00% and a Dice coefficient of 95.25% on a dataset of 70 radiographs.
• For images with high bone superposition, the average detection rate was 96.36% and the Dice coefficient was 93.81%.

Conclusions:

• The proposed method effectively segments lower limb bones, including overlapping structures and partial bone fragments.
• This automated approach demonstrates significant potential for clinical applications in orthopedics.