Automatic lung tumor segmentation on PET/CT images using fuzzy Markov random field model

Area of Science:

• Medical Imaging
• Radiology
• Computational Biology

Background:

• Positron Emission Tomography (PET) and Computed Tomography (CT) imaging fusion offers complementary functional and anatomical data for improved lung cancer tumor volume definition.
• Accurate segmentation of lung tumors, especially those adjacent to tissues with similar image intensities, remains a challenge in clinical practice.

Purpose of the Study:

• To develop and evaluate a robust, automated method for lung tumor segmentation using PET/CT images.
• To improve the accuracy and efficiency of defining tumor boundaries for non-small cell lung cancer (NSCLC) patients.

Main Methods:

• A novel fuzzy Markov random field (MRF) model was employed for automatic tumor segmentation.
• The method utilizes a specialized joint posterior probability distribution to effectively combine PET and CT image information, outperforming standard Gaussian distributions.
• The algorithm was validated using simulated PET/CT images from 7 NSCLC patients.

Main Results:

• The proposed fuzzy MRF method achieved high accuracy in segmenting lung tumors, with a Dice's similarity coefficient (DSC) of 0.85 ± 0.013 when compared to manual segmentation by an experienced radiation oncologist.
• The method demonstrated effectiveness in segmenting tumors located near organs with similar intensities, such as those extending into the chest wall or mediastinum.

Conclusions:

• The developed fuzzy MRF-based method provides an effective and automatic solution for lung tumor segmentation on PET/CT images.
• This automated approach shows significant potential for clinical application in precise tumor delineation for NSCLC treatment planning.