Improving Data-Efficiency and Robustness of Medical Imaging Segmentation Using Inpainting-Based Self-Supervised Learning.
Improving Data-Efficiency and Robustness of Medical Imaging Segmentation Using Inpainting-Based Self-Supervised Learning.
复制标题
使用基于介入的自我监督学习来改善医学成像分割的数据效率和鲁棒性。
DOI：
10.3390/bioengineering10020207
复制
发表时间：
2023-02-04
期刊：
影响因子：
--
通讯作者：
中科院分区：
文献类型：
--
作者：
我们系统地评估训练方法和功效的两个inpainting为基础的借口任务的上下文预测和上下文恢复的医学图像分割使用自监督学习（SSL）。训练多个版本的自监督U-Net模型来分割MRI和CT数据集，每个模型使用不同的设计选择和借口任务组合来确定这些设计选择对分割性能的影响。最佳设计选择用于训练SSL模型，然后将其与基线监督模型进行比较，以计算标签限制场景中的临床相关指标。我们观察到，SSL预训练使用32 × 32补丁和Poission-disc采样进行上下文恢复，仅传输预训练的编码器权重，并以1 × 10的初始学习率立即进行微调，这比监督学习对MRI和CT组织分割准确性的益处最大（p < 0.001）。对于数据集和大多数标签受限的场景，缩放未标记的预训练数据的大小可以提高分割性能。用这些数据量预训练的SSL模型在计算临床相关指标方面优于基线监督模型，特别是当监督学习的性能较低时。我们的研究结果表明，使用基于inpainting的借口任务的SSL预训练可以帮助提高模型在标签受限场景中的鲁棒性，并减少监督学习中发生的最坏情况错误。
We systematically evaluate the training methodology and efficacy of two inpainting-based pretext tasks of context prediction and context restoration for medical image segmentation using self-supervised learning (SSL). Multiple versions of self-supervised U-Net models were trained to segment MRI and CT datasets, each using a different combination of design choices and pretext tasks to determine the effect of these design choices on segmentation performance. The optimal design choices were used to train SSL models that were then compared with baseline supervised models for computing clinically-relevant metrics in label-limited scenarios. We observed that SSL pretraining with context restoration using 32 × 32 patches and Poission-disc sampling, transferring only the pretrained encoder weights, and fine-tuning immediately with an initial learning rate of 1 × 10 provided the most benefit over supervised learning for MRI and CT tissue segmentation accuracy (p < 0.001). For both datasets and most label-limited scenarios, scaling the size of unlabeled pretraining data resulted in improved segmentation performance. SSL models pretrained with this amount of data outperformed baseline supervised models in the computation of clinically-relevant metrics, especially when the performance of supervised learning was low. Our results demonstrate that SSL pretraining using inpainting-based pretext tasks can help increase the robustness of models in label-limited scenarios and reduce worst-case errors that occur with supervised learning.
登录
查看更多内容
DOI：
10.2214/ajr.20.24172
发表时间：
2021-06
期刊：
AJR. American journal of roentgenology
影响因子：
--
作者：
Chaudhari AS;Grissom MJ;Fang Z;Sveinsson B;Lee JH;Gold GE;Hargreaves BA;Stevens KJ
通讯作者：
Stevens KJ
DOI：
10.1002/jmri.26991
发表时间：
2020-11
期刊：
Journal of magnetic resonance imaging : JMRI
影响因子：
--
作者：
Chaudhari AS;Kogan F;Pedoia V;Majumdar S;Gold GE;Hargreaves BA
通讯作者：
Hargreaves BA
影响因子：
4.4
作者：
Chaudhari, Akshay S.;Stevens, Kathryn J.;Hargreaves, Brian A.
通讯作者：
Hargreaves, Brian A.
影响因子：
1.9
作者：
Boutin, Robert D.;Houston, Denise K.;Lenchik, Leon
通讯作者：
Lenchik, Leon
影响因子：
5.9
作者：
Eijgenraam, Susanne M.;Chaudhari, Akshay S.;Oei, Edwin H. G.
通讯作者：
Oei, Edwin H. G.