Abstract:Medical image segmentation is a fundamental yet challenging task due to the arduous process of acquiring large volumes of high-quality labeled data from experts. Contrastive learning offers a promising but still problematic solution to this dilemma. Because existing medical contrastive learning strategies focus on extracting image-level representation, which ignores abundant multi-level representations. And they underutilize the decoder either by random initialization or separate pre-training from the encoder, thereby neglecting the potential collaboration between the encoder and decoder. To address these issues, we propose a novel multi-level asymmetric contrastive learning framework named MACL for volumetric medical image segmentation pre-training. Specifically, we design an asymmetric contrastive learning structure to pre-train encoder and decoder simultaneously to provide better initialization for segmentation models. Moreover, we develop a multi-level contrastive learning strategy that integrates correspondences across feature-level, image-level, and pixel-level representations to ensure the encoder and decoder capture comprehensive details from representations of varying scales and granularities during the pre-training phase. Finally, experiments on 12 volumetric medical image datasets indicate our MACL framework outperforms existing 11 contrastive learning strategies. {\itshape i.e.} Our MACL achieves a superior performance with more precise predictions from visualization figures and 2.28\%, 1.32\%, 1.62\% and 1.60\% Average Dice higher than previous best results on CHD, MMWHS, CHAOS and AMOS, respectively. And our MACL also has a strong generalization ability among 5 variant U-Net backbones. Our code will be available at <a class="link-external link-https" href="https://github.com/stevezs315/MACL" rel="external noopener nofollow">this https URL</a>.

LACL: Lesion-Aware Contrastive Learning Framework for Medical Image Classification

Combining Convolutional and Recurrent Neural Networks for Classification of Focal Liver Lesions in Multi-phase CT Images.

A Clinical-oriented Multi-level Contrastive Learning Method for Disease Diagnosis in Low-quality Medical Images

Lesion-Aware Contrastive Representation Learning for Histopathology Whole Slide Images Analysis

Lesion-Aware Contrastive Learning for Diabetic Retinopathy Diagnosis

TL-CCL: Two-level causal contrastive learning for multi-label ocular disease diagnosis with fundus images

Lesion-based Contrastive Learning for Diabetic Retinopathy Grading from Fundus Images

Multi-level Asymmetric Contrastive Learning for Medical Image Segmentation Pre-training

A Cascade Attention Network for Liver Lesion Classification in Weakly-Labeled Multi-phase CT Images.

Multi-level Asymmetric Contrastive Learning for Volumetric Medical Image Segmentation Pre-training

Fine-grained Lesion Classification Framework for Early Auxiliary Diagnosis

Positional Contrastive Learning for Volumetric Medical Image Segmentation

Local Contrastive Learning for Medical Image Recognition

Unsupervised Local Discrimination for Medical Images

CCAC: Contrastive Learning with Channel Attention and Contour Loss for Brain Image Registration.

Contrastive Centroid Supervision Alleviates Domain Shift in Medical Image Classification

Non-Local Attention Learning for Medical Image Classification

Class-aware Patch Based Contrastive Learning for Medical Image Segmentation

Triplet attention and dual-pool contrastive learning for clinic-driven multi-label medical image classification

Anatomy-Aware Contrastive Representation Learning for Fetal Ultrasound