Deep learning?based multi?modal computing with feature disentanglement for MRI image synthesis

Purpose: Different Magnetic resonance imaging (MRI) modalities of the same anatomical structure are required to present different pathological information from physical level for diagnostic needs. However, it is often difficult obtain full-sequence MRI images patients owing limitations such as time consumption and high cost. The purpose this work develop an algorithm target sequences prediction with accuracy, provide more clinical diagnosis. Methods: We propose a deep learning based multi-modal computing model synthesis feature disentanglement strategy. To take full advantage complementary provided by modalities, utilized input. Notably, proposed approach decomposes each input modality into modality-invariant space shared modality-specific specific information, so that features extracted separately effectively process data. Subsequently, both them fused through adaptive instance normalization (AdaIN) layer in decoder. In addition, address lack test phase, local fusion (LAF) module adopted generate modality-like pseudo-target similar ground truth. Results: evaluate performance, we verify our method on BRATS2015 dataset 164 subjects. experimental results demonstrate significantly outperforms benchmark other state-of-the-art medical image methods quantitative qualitative measures. Compared pix2pixGANs method, PSNR improves 23.68 24.8. Conclusion: could be effective sequences, useful diagnosis treatment.