Deep learning classifiers face significant challenges when dealing with
heterogeneous multi-modal and multi-organ biomedical datasets. The low-level
feature distinguishability limited to imaging-modality hinders the classifiers'
ability to learn high-level semantic relationships, resulting in sub-optimal
performance. To address this issue, image augmentation strategies are employed
as regularization techniques. While additive noise input during network
training is a well-established augmentation as regularization method, modern
pipelines often favor more robust techniques such as dropout and weight decay.
This preference stems from the observation that combining these established
techniques with noise input can adversely affect model performance.
In this study, we propose a novel pretraining pipeline that learns to
generate conditional noise mask specifically tailored to improve performance on
multi-modal and multi-organ datasets. As a reinforcement learning algorithm,
our approach employs a dual-component system comprising a very light-weight
policy network that learns to sample conditional noise using a differentiable
beta distribution and a classifier network. The policy network is trained using
the reinforce algorithm to generate image-specific noise masks that regularize
the classifier during pretraining. A key aspect is that the policy network's
role is limited to obtaining an intermediate (or heated) model before
fine-tuning. During inference, the policy network is omitted, allowing direct
comparison between the baseline and noise-regularized models.
We conducted experiments and related analyses on RadImageNet datasets.
Results demonstrate that fine-tuning the intermediate models consistently
outperforms conventional training algorithms on both classification and
generalization to unseen concept tasks.

本研究提出了一种学习生成特定噪声遮罩的预训练流程，旨在改善在多模态和多器官数据集上的性能，实验证明中间模型的微调能够在分类和概念任务上持续优于传统训练算法。

政策梯度驱动的噪声遮罩

Policy Gradient-Driven Noise Mask

Self-supervised learning has proven to be an effective way to learn
representations in domains where annotated labels are scarce, such as medical
imaging. A widely adopted framework for this purpose is contrastive learning
and it has been applied to different scenarios. This paper seeks to advance our
understanding of the contrastive learning framework by exploring a novel
perspective: employing multi-organ datasets for pre-training models tailored to
specific organ-related target tasks. More specifically, our target task is
breast tumour segmentation in ultrasound images. The pre-training datasets
include ultrasound images from other organs, such as the lungs and heart, and
large datasets of natural images. Our results show that conventional
contrastive learning pre-training improves performance compared to supervised
baseline approaches. Furthermore, our pre-trained models achieve comparable
performance when fine-tuned with only half of the available labelled data. Our
findings also show the advantages of pre-training on diverse organ data for
improving performance in the downstream task.

利用自我监督学习和对比学习的方法，在医学成像等标注稀缺的领域中学习表征已被证明是一种有效的方式。本研究通过探索一种新的思路 —— 使用多器官数据集来为特定器官相关目标任务预训练模型，旨在推进我们对对比学习框架的理解。具体而言，我们的目标任务是超声图像中的乳腺肿瘤分割。预训练数据集包括来自其他器官（如肺和心脏）的超声图像以及大量的自然图像。我们的结果表明，与监督基线方法相比，传统的对比学习预训练可以提高性能。此外，我们的预训练模型在只使用一半可用标记数据进行微调时也可以达到相当的性能。我们的发现还表明，预训练多器官数据可以提高下游任务的性能。