Computerized scar detection on renal cortical scintigraphy images.

Area of Science:

• Nuclear Medicine
• Medical Imaging
• Computer-Aided Diagnosis

Background:

• Renal cortical scintigraphy is a functional imaging technique.
• Visual evaluation of scintigraphy is subjective, leading to interobserver variability in scar detection.
• Quantifying renal scars requires objective assessment methods.

Purpose of the Study:

• To develop novel computerized methods for renal cortical scintigraphy image interpretation.
• To address challenges in analyzing radiopharmaceutical tracer activity distribution and cortex continuity (scars).
• To reduce subjectivity and interobserver variability in scar identification.

Main Methods:

• Implemented model-based automatic kidney segmentation using active-shape models.
• Utilized image normalization to standardize kidney images into vectors.
• Introduced new image-based features: localized activity drop and Principal Component Analysis (PCA).
• Evaluated performance using free-response receiver operating characteristic (FROC) analysis.

Main Results:

• The PCA-based feature demonstrated the highest scar detection performance.
• Localized activity drop feature also showed improved performance compared to previous schemes.
• PCA-based scheme achieved a sensitivity of 0.90 with a mean of 4.52 false positives (FPs) per image.
• Adjusting sensitivity to 0.79 improved the mean FPs to 1.21.
• A decision fusion scheme achieved 0.83 sensitivity with 1.90 mean FPs.

Conclusions:

• The developed computerized methods offer potential as a second reader for nuclear medicine specialists.
• These methods can assist in identifying renal scar regions more objectively.
• Further improvements to reduce the false positive rate are being explored.