Single-Channel Sparse Non-Negative Blind Source Separation Method for Automatic 3-D Delineation of Lung Tumor in PET...

Area of Science:

• Medical Imaging
• Signal Processing
• Oncology

Background:

• Accurate lung tumor segmentation in Positron Emission Tomography (PET) is crucial for cancer treatment.
• Current segmentation methods face challenges due to image noise and tumor heterogeneity.
• Automated segmentation can improve efficiency and consistency in clinical practice.

Purpose of the Study:

• To introduce a novel, fully automated method for lung tumor segmentation in PET images.
• To evaluate the performance of the proposed method against existing algorithms and ground truth.
• To assess the potential clinical applicability of the developed technique.

Main Methods:

• A novel single-channel blind source separation technique was adapted for PET image analysis.
• 3D PET images were converted into pseudo-multichannel images.
• Regularization-free, sparseness-constrained non-negative matrix factorization (NMF) was employed for tissue separation.
• A complexity-based criterion was used to identify the tumor component.
• The method was compared against thresholding, graph cuts (GC), random walks (RW), and affinity propagation (AP) on 18 non-small cell lung cancer datasets.

Main Results:

• The proposed algorithm achieved a Dice similarity coefficient (DSC) of 0.78 ± 0.12.
• Performance was comparable to established methods like GC (0.78 ± 0.1) and AP (0.77 ± 0.13).
• The method demonstrated competitive results against RW (0.77 ± 0.07) and 50% SUV thresholding (0.75 ± 0.13).

Conclusions:

• The proposed automated method provides accurate lung tumor segmentation in PET images.
• Its performance is comparable to interactive segmentation techniques.
• The findings support the potential use of this fully automated method in routine clinical settings for lung cancer management.