New strategy for automatic tumor segmentation by adaptive thresholding on PET/CT images

Area of Science:

• Nuclear Medicine
• Radiotherapy
• Medical Imaging Analysis

Background:

• Accurate tumor delineation is crucial for radiotherapy planning, traditionally relying on computed tomography (CT) anatomical imaging.
• Positron emission tomography (PET) functional imaging is recommended for non-small cell lung cancer to capture biological characteristics.
• Current PET image segmentation techniques lack satisfactory clinical application performance.

Purpose of the Study:

• To develop and validate a novel segmentation technique for positron emission tomography (PET) images in clinical settings.
• To address the limitations of existing segmentation methods for non-small cell lung cancer treatment planning.

Main Methods:

• A new segmentation technique was developed using patient-specific threshold adjustments, referencing CT and histological measurements.
• Sixty-five lung lesions from 54 patients undergoing FDG-PET/CT scans were analyzed.
• Segmentation accuracy was validated by comparing PET-derived measurements with CT and histological data.

Main Results:

• The novel technique demonstrated good agreement with histological measurements (-0.8 ± 9.0% difference) and acceptable CT estimation for lesions >20 mm.
• For lesions ≤20 mm, the technique showed disagreement with histological and CT measurements.
• High accuracy was observed for lesions with largest axes between 2 and 4.5 cm, with limitations for smaller lesions potentially due to partial volume effects or motion.

Conclusions:

• The proposed PET image segmentation technique offers high accuracy for delineating larger non-small cell lung cancer lesions.
• The method's effectiveness is limited for smaller lesions, highlighting challenges related to partial volume effects and respiratory motion.
• This technique shows promise for improving radiotherapy planning by enhancing the delineation of significant tumor volumes on PET images.