In the first study of its kind, researchers used positron emission tomography and computed tomography (PET-CT) feature fusion for segmenting diffuse large B-cell lymphoma (DLBCL). The findings were published in the journal Medical Physics.
To conduct this analysis, researchers first extracted single-modality features using encoder branches, before using the learning method to generate spatial fusion maps. Subsequently, the fusion maps were linked with specific-modality maps to derive a representation of final-fused maps in different scales. Lastly, a reconstruction component produced a prediction map of DLBCL lesions by integrating and up-sampling the final-fused maps. The method was used to detect both foreground and segment lesions in the nasopharynx, chest, and abdomen of 45 PET-CT scans.
According to the results, the method achieved a high detection accuracy in the nasopharynx (99.63%), chest (99.51%), and abdomen (99.21%), and exhibited superiority in segmentation performance.
“A promising segmentation method has been proposed for the challenging DLBCL lesions in PET-CT images, which improves the understanding of complementary information by feature fusion and may guide clinical radiotherapy,” the researchers concluded.
They added that “this is a preliminary research using a small sample size, and we will collect data continually to achieve the larger verification study.”