Modality-Specific Segmentation Network for Lung Tumor Segmentation in PET-CT Images.

Area of Science:

• Medical imaging
• Artificial intelligence in oncology
• Radiomics

Background:

• Accurate lung tumor segmentation in PET-CT images is crucial for cancer diagnosis and treatment planning.
• Tumor appearance variability between PET and CT modalities poses a significant segmentation challenge due to patient movement and respiration.
• Existing methods struggle to effectively address the modality-specific differences in lung tumor representation.

Purpose of the Study:

• To propose a novel modality-specific segmentation network (MoSNet) for accurate lung tumor segmentation in PET-CT images.
• To develop a method that simultaneously segments tumors in PET and CT images while accounting for inter-modality inconsistencies.
• To enhance the representation power for modality-specific lung tumor segmentation.

Main Methods:

• Developed MoSNet, a network that learns modality-specific and modality-fused representations of lung tumors.
• Employed an adversarial method with a modality discriminator to minimize modality discrepancy and preserve common features.
• Introduced a modality-specific map to quantify feature weights for each modality.

Main Results:

• MoSNet demonstrated superior performance in segmenting lung tumors across 126 PET-CT images of non-small cell lung cancer (NSCLC).
• The proposed method effectively handles the shape and size variations of lung tumors between PET and CT images.
• Experimental results confirmed MoSNet's outperformance compared to state-of-the-art segmentation techniques.

Conclusions:

• MoSNet offers a significant advancement in lung tumor segmentation for PET-CT imaging.
• The network's ability to learn and leverage modality-specific features improves segmentation accuracy.
• This approach holds promise for enhancing clinical decision-making in lung cancer management.