Fuzzy hidden Markov chains segmentation for volume determination and quantitation in PET.

Area of Science:

• Medical Imaging
• Computational Biology
• Radiology

Background:

• Accurate volume of interest (VOI) estimation is critical for oncology applications like therapy response evaluation and radiotherapy planning.
• Current threshold-based techniques for lesion delineation in PET imaging have limitations in precision and noise susceptibility.

Purpose of the Study:

• To evaluate the performance of a novel fuzzy hidden Markov chains (FHMC) algorithm for automatic lesion volume delineation.
• To compare FHMC against state-of-the-art threshold-based methods in positron emission tomography (PET) data.

Main Methods:

• FHMC algorithm incorporates noise, voxel intensity, and spatial correlation, with a novel fuzzy model estimating imprecision for better boundary modeling.
• Performance assessed on simulated and acquired IEC phantom datasets with varying lesion sizes, contrast ratios, and noise levels.
• Evaluation metrics included lesion activity recovery, VOI determination, and % classification errors.

Main Results:

• FHMC significantly outperformed threshold-based methods for functional volume determination and activity concentration recovery, particularly for lesions smaller than 28 mm at a 4:1 contrast ratio.
• FHMC demonstrated smaller classification and volume estimation errors.
• FHMC exhibited greater robustness against image noise compared to threshold-based techniques.

Conclusions:

• The FHMC algorithm provides a more accurate and robust approach for automatic lesion volume delineation in PET imaging.
• FHMC enhances the reliability of functional VOI estimation, crucial for clinical oncology applications.
• The fuzzy modeling in FHMC effectively addresses the inherent imprecision in emission tomography data boundaries.