A Multiprocessing Scheme for PET Image Pre-Screening, Noise Reduction, Segmentation and Lesion Partitioning

Area of Science:

• Medical Imaging
• Computational Biology
• Radiology

Background:

• Manual segmentation and partitioning of lesions in PET images are time-consuming, laborious, and prone to inaccuracies.
• Accurate lesion characterization is crucial for tumor diagnosis, staging, and prognosis.

Purpose of the Study:

• To develop an automated, multiprocessing scheme for PET image pre-screening, noise reduction, segmentation, and lesion partitioning.
• To enhance the accuracy and efficiency of lesion analysis in PET imaging.

Main Methods:

• Proposed a differential activation filter (DAF) for efficient lesion pre-screening in whole-body PET scans.
• Introduced a neural network inverse (NN inverse) of generalized Anscombe transformation (GAT) for robust noise reduction, improving signal-to-noise ratio (SNR).
• Developed definition density peak clustering (DDPC) for unsupervised instance segmentation of lesions and normal tissues, reducing computational cost.

Main Results:

• The automated scheme significantly reduces the time required for PET image analysis.
• The proposed methods demonstrate superior performance in noise reduction and segmentation accuracy compared to existing state-of-the-art techniques.
• Experimental results on clinical data validate the effectiveness of the DAF, NN inverse, and DDPC methods.

Conclusions:

• The developed automated multiprocessing scheme offers a more efficient and accurate alternative to manual lesion analysis in PET imaging.
• This approach has the potential to improve computer-aided diagnosis and clinical decision-making in oncology.
• The study highlights the benefits of integrating advanced image processing techniques for enhanced PET image analysis.