在醫學圖象中，多模態數據因成像機理不一樣而能從多種層面提供信息。多模態圖像分割包含重點問題爲如何融合(fusion)不一樣模態間信息，本文主要記錄筆者最近所讀，歡迎批評指正補充
1. A review: Deep learning for medical image segmentation using multi-modality fusion (Array 2019）***
融合策略分類
綜述，按照方法的位置將融合策略分爲三大類：Input-level, Layer-level, Decision-level. spa
數據集：3d
幾種多模態數據集
2. Co-Learning Feature Fusion Maps from PET-CT Images of Lung Cancer (TMI 2019）****
Abstract : Layer-level fusion, U-net, PET-CT, Lung cancercode
Method : 對CT與PET各一個encoder，不一樣層feature stack後經conv獲得權重，與concat後的feature點積，得加權feature map。blog
Experiment : 肺癌數據，對比layer-level fusion : MB(multi-branch), MC(multi-channel), FS(fused)，效果不錯。ci
3. 3D FULLY CONVOLUTIONAL NETWORKS FOR CO-SEGMENTATION OF TUMORS ON PET-CT IMAGES (ISBI 2018）**
Abstract : Decision-level, Unet, Graph cut, PET-CT, Lung cancerit
Method : CT, PET獨立的Net，各輸出機率圖後Graph Cut。io
Experiment : 肺癌數據，對比graph-cut。class
4. HyperDense-Net: A hyper-densely connected CNN for multi-modal image segmentation (TMI 2019) ****
Abstract : Layer-level, DenseNet, MRI, Brainsed
Method : modality各一個net, 中間層相互dense鏈接。network
Experiment : Brain(iseg-2017, MRBrainS)，對比layer level fusion : Single Dense Path, Dual Dense Path, Disentangled modalities with early fusion。各模態先經卷積再拼接相比直接雙通道輸入有較明顯提高。
5. Deep learning for automatic tumour segmentation in PET/CT images of patients with head and neck cancers (MIDL 2019) *
Abstract : Input-level, Unet, Pet-CT, head and neck
Method : 2通道輸入
Experiment :頭頸部腫瘤， 對比單模態Unet，是HECKTOR比賽數據來源。
6. Tumor co-segmentation in PET/CT using multi-modality fully convolutional neural network (Physics in Medicine & Biology 2019) **
Abstract : Layer-level, V-net, Pet-CT, Lung cancer
Method : 2個V-net 先分別提取PET/CT feature，sum後經4層卷積得result. 提出weighted cross entropy loss以balance不一樣模態影響。
Experiment : 肺癌數據，對比其它幾種fusion方法，傳統方法，單模態V-net。
歡迎評論與補充相關的論文～