Computerized segmentation and diagnostics of whole-body bone scintigrams

Area of Science:

• Nuclear Medicine
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Bone scintigraphy (whole-body bone scan) is a common nuclear medicine diagnostic procedure.
• Manual image evaluation by physicians is time-consuming and subjective.
• A need exists for automated methods to improve efficiency and accuracy in pathology detection.

Purpose of the Study:

• To present a knowledge-based methodology for segmenting whole-body bone scans into skeletal regions.
• To develop an automated system for pathology detection and classification in bone scintigraphy.
• To compare the performance of the automated system against expert physician evaluations.

Main Methods:

• A knowledge-based methodology was developed for segmenting anterior and posterior whole-body bone scintigrams.
• Expert knowledge was encoded using parameterized rules to guide image processing algorithms.
• Segmented regions were parameterized for pattern classification using machine learning algorithms.

Main Results:

• The automated segmentation algorithm was applied to 467 consecutive bone scintigrams.
• The system demonstrated more accurate and reliable results compared to previous studies.
• Preliminary experiments indicated that the machine learning-based expert system closely replicated expert physician findings.

Conclusions:

• The developed knowledge-based segmentation methodology enables automatic evaluation of pathological changes in bone scans.
• This automated approach can detect not only point-like lesions but also other types of pathologies.
• The system shows potential to assist physicians in nuclear medicine diagnostics, improving accuracy and reliability.