Abstract A comprehensive segmentation of brain tumors in tomographic image data requires multiple MRI sequences. However, some sequence modalities may be unavailable due to clinical constraints or artifacts. This study investigates the impact of using synthetic contrast-enhanced T1 images (ceT1) on the automatic tumor segmentation performance. A generative model based on the Pix2Pix architecture was trained to synthesize ceT1 from native T1, T2, and FLAIR images. A segmentation model was then trained and tested using various substitution strategies for missing ceT1, including synthetic images, copied native T1, and empty inputs. The synthetic ceT1 achieved an average structural similarity index of 92.6% and a peak-signal-to-noise ratio of 21.918 dB. When used in segmentation, the synthetic modality maintained near-baseline performance for whole tumor and tumor core regions. However, segmentation accuracy for enhancing tumor regions dropped substantially due to the absence of true contrast-agent uptake. In conclusion, synthetic ceT1 can effectively replace missing images for segmenting whole tumor and tumor core regions, outperforming naive alternatives. Nevertheless, its use remains limited for enhancing tumor detection, which depends on physiological contrast accumulation not captured by image synthesis.
Rothert et al. (Mon,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: