Automated bony region identification using artificial neural networks: reliability and validation measurements

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Orthopedic research

Background:

• Manual segmentation of bony structures from medical images is time-consuming and prone to variability.
• Accurate identification of phalanges is crucial for orthopedic assessments and patient-specific modeling.

Purpose of the Study:

• To develop an automated tool for identifying bony structures using artificial intelligence.
• To evaluate the reliability and accuracy of this automated technique against manual methods and physical scans.
• To assess the efficiency of the automated approach in terms of time and effort.

Main Methods:

• Development and training of an artificial neural network (ANN) for image segmentation.
• ANN was trained to specifically identify the phalanges of the human hand from CT images.
• Comparison of ANN segmentation with manual tracing and physical surface scans.

Main Results:

• The ANN demonstrated high overlap with manual tracings (0.76–0.87) for index phalanx bones.
• ANN-generated surfaces showed minimal average differences (0.29–0.40 mm) compared to physical scans.
• The ANN segmented structures significantly faster, in less than one-tenth the time of manual raters.

Conclusions:

• Artificial neural networks provide a reliable and valid method for segmenting phalanx bones from CT images.
• Automated segmentation eliminates rater drift and inter-rater variability, improving consistency.
• This automation reduces time and effort, facilitating patient-specific modeling.