A novel fuzzy C-means algorithm for unsupervised heterogeneous tumor quantification in PET.

Area of Science:

• Medical Imaging
• Radiology
• Computational Biology

Background:

• Accurate PET quantification in oncology is challenged by low spatial resolution and high noise.
• Standard fuzzy C-means (FCM) segmentation is sensitive to noise and intensity variations.
• Existing methods lack spatial contextual information crucial for accurate segmentation.

Purpose of the Study:

• To develop a novel fuzzy segmentation technique for noisy, low-resolution oncological PET data.
• To enhance FCM by incorporating spatial information and wavelet transforms for improved tumor segmentation.
• To accurately quantify tumor size and volume in non-small-cell lung cancer (NSCLC) and laryngeal squamous cell carcinoma (LSCC) patients.

Main Methods:

• Proposed a new FCM algorithm integrating smoothed PET images (FCM-S) for spatial information.
• Developed FCM with a trous wavelet transform (FCM-SW) to handle heterogeneous tracer uptake.
• Validated the algorithm on simulated NCAT phantom data and clinical PET/CT scans from NSCLC and LSCC patients.

Main Results:

• FCM-SW demonstrated a strong correlation (R2=0.942) between estimated and actual maximal tumor diameters in NSCLC.
• Reduced mean error in maximal diameter estimation from -4.6 mm (FCM) to 0.1 mm (FCM-SW).
• Decreased mean relative volume error from 21.7% (FCM) to 8.6% (FCM-SW) for LSCC patients.

Conclusions:

• A novel unsupervised PET segmentation technique accurately quantifies lesions with heterogeneous uptake.
• The FCM-SW algorithm improves tumor size and volume estimation in oncological PET imaging.
• The technique shows promise for PET-guided radiation therapy and treatment response assessment.