Semi-supervised cardiac MRI image of the left ventricle segmentation algorithm based on contrastive learning

Abstract

A semi-supervised convolutional neural network segmentation method of medical images based on contrastive learning is proposed. The cardiac magnetic resonance imaging (MRI) images to be segmented are preprocessed to obtain positive and negative samples by labels. The U-Net shrinks network is applied to extract features of the positive samples, negative samples, and input samples. In addition, an unbalanced contrastive loss function is proposed, which is weighted with the binary cross-entropy loss function to obtain the total loss function. The model is pre-trained with labeled samples, and unlabeled images are predicted by the pre-trained model to generate pseudo-labels. A pseudo-label post-processing algorithm for removing disconnected regions and hole filling of pseudo-labels is proposed to guide the training process of semi-supervised networks. The results on the Sunnybrook dataset show that the segmentation results of this model are better, with a higher dice coefficient, accuracy, and recall rate.

References

[1] VhadsellR, ChopraS, LecunY. Dimensionality reduction by learning an invariant mapping[C], 2006, New York, IEEE: 1735-17422

[2] HeK, FanH, WuY, et al.. Momentum contrast for unsupervised visual representation learning[C], 2020, New York, IEEE: 9729-9738

[3] ChenT, KornblithS, NorouziM, et al.. A simple framework for contrastive learning of visual representa-tions[C], 2020, San Diego, ICML: 1597-1607

[4] LeeD H. Pseudo-label: the simple and efficient semi-supervised learning method for deep neural net-works[C], 2013, San Diego, ICML: 896

[5] KhoslaP, TeterwakP, WangC, et al.. Supervised contrastive learning[J]. Advances in neural information processing systems, 2020, 33: 18661-18673

[6] ChaitanyaK, ErdilE, KaraniN, et al.. Contras-tive learning of global and local features for medical image segmentation with limited annotations[J]. Advances in neural information processing systems, 2020, 33: 12546-12558

[7] ZHENG X, FU C, XIE H, et al. Uncertainty-aware deep co-training for semi-supervised medical image segmen-tation[EB/OL]. (2021-11-23) [2022-04-26]. https://arxiv.org/abs/2111.11629v1.

[8] ChakrabortyS, GosthipatyA R, PaulS. G-SimCLR: self-supervised contrastive learning with guided projection via pseudo labelling[C], 2020, New York, IEEE: 912-916

[9] DIPPEL J, VOGLER S, HÖHNE J. Towards fine-grained visual representations by combining con-trastive learning with image reconstruction and attention-weighted pooling[EB/OL]. (2021-04-09) [2022-04-26]. https://arxiv.org/abs/2104.04323.

[10] HeK, ZhangX, RenS, et al.. Delving deep into rectifiers: surpassing human-level performance on Imagenet classification[C], 2015, New York, IEEE: 1026-1034

[11] ZHAO X, FANG C, FAN D J, et al. Cross-level con-trastive learning and consistency constraint for semi-supervised medical image segmentation[EB/OL]. (2021-02-13) [2022-04-26]. https://arxiv.org/abs/2202.04074.