自适应损失函数提升胰腺医学图像分割精度
UNet-3D with Adaptive TverskyCE Loss for Pancreas Medical Image Segmentation
- 设计可学习权重的混合损失函数，动态调整分割优化目标
- 在NIH数据集上达85.59%的Dice系数，最高95.24%
- 适合医疗影像分割研究者与临床辅助诊断系统开发者
胰腺癌生存率低，早期诊断依赖腹部计算机断层扫描（CT），但因胰腺位置隐蔽、体积小且常被周围器官遮挡，分割难度大。尽管深度学习模型在分割任务中表现良好，仍需提升性能以应对挑战。本文提出一种自适应TverskyCE损失函数，通过可学习权重将Tversky损失与交叉熵损失融合，实现训练过程中损失贡献的自动调节，动态优化目标函数。所有实验基于美国国立卫生研究院（NIH）胰腺CT数据集进行，评估了UNet-3D与Dilated UNet-3D模型。所提方法在胰腺分割任务中取得85.59%的骰子相似系数（DSC），峰值达95.24%，平均得分85.14%，相比基线UNet-3D加Tversky损失，分别提升9.47%和8.98%。
原文摘要 · Abstract (English)
Pancreatic cancer, which has a low survival rate, is one of the most challenging cancers to diagnose and treat effectively. Early detection through abdominal computed tomography (CT) scans is crucial, yet complicated by the pancreas' obscure anatomical position, small size, and frequent occlusion by surrounding organs. These factors make the pancreas particularly difficult to identify and segment accurately. While deep learning (DL) models have shown promise for segmentation tasks, their performance still requires significant improvement to address these challenges. In this research, we propose a novel adaptive TverskyCE loss for DL model training, which combines Tversky loss with cross-entropy loss through learnable weights. Our method enables automatic adjustment of loss contributions during training, dynamically optimizing the objective function for improved performance. All experiments were conducted on the National Institutes of Health (NIH) Pancreas-CT dataset. We evaluated the adaptive TverskyCE loss on the UNet-3D and Dilated UNet-3D, and our method achieved a Dice Similarity Coefficient (DSC) of 85.59%, with peak performance up to 95.24%, and the score of 85.14%. DSC and the score score were improved by 9.47% and 8.98% respectively compared with the baseline UNet-3D with Tversky loss for pancreas segmentation. Keywords: Pancreas segmentation, Tversky loss, Cross-entropy loss, UNet-3D, Dilated UNet-3D
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。