Zhang Y, Zhou T, Wang S, et al. Input augmentation with sam: Boosting medical image segmentation with segmentation foundation model[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer Nature Switzerland, 2023: 129-139.【开源】
这篇文章想法比较好,既然最近的工作表明,单独使用 SAM,如果没有进一步的微调和域适应,通常无法为医学图像分割任务提供令人满意的结果(在后面几篇文章中都提到了这个观点), 你就将其用于提供更好的先验信息。这里具体解读一下:
【论文概述】
本文提出了一种名为SAMAug的新方法,以提升医学图像分割的效果。这种方法利用了一个称为Segment Anything Model (SAM)的大型基础模型,通过增强医学图像输入来改进常用的医学图像分割模型。尽管SAM在医学图像数据上直接应用时并不立即产生高质量的分割效果,但它生成的掩膜特征和稳定性评分对于构建和训练更好的医学图像分割模型非常有用。文中通过在三个分割任务上的实验,展示了SAMAug方法的有效性。这项工作的主要贡献包括:利用SAM提供的分割输出与原始图像输入的结合,以及为医学图像分割模型构建SAM增强输入图像。此外,还对SAM在医学图像分析中的潜力进行了进一步探索。
【方法】
总的来说,这个方法通过利用SAM生成的先验信息,增强医学图像的输入数据,以改进医学图像分割模型的训练和性能。
【实验结果】
可以看出经过这种方式处理,涨点 很明显
Huang Y, Yang X, Liu L, et al. Segment anything model for medical images?[J]. Medical Image Analysis, 2023: 103061.[【开源】](https://github.com/yuhoo0302/
Segment-Anything-Model-for-Medical-Images.)
本文是直接实验验证,没有改变SAM,不过多解读,只关注结论
为了全面验证SAM在医学数据上的表现,收集并整理了53个开源数据集,并建立了一个大型医学分割数据集,包含18种成像模式、84种物体、125个物体-模式配对目标、1050K 2D图像和6033K掩膜。对所谓的COSMOS 1050K数据集上的不同模型和策略进行了全面分析。发现主要包括
SAM在某些特定物体上表现出色,但在其他情况下表现不稳定、不完美甚至完全失败。
搭载大型ViT-H的SAM总体性能优于搭载小型ViT-B的。
SAM在手动提示(尤其是框提示)下的表现优于"Everything"模式。
SAM可以帮助人类进行高质量、节省时间的标注。SAM对中心点和紧凑框提示的随机性敏感,可能会导致严重的性能下降。
SAM比只用一个或几个点的交互式方法表现更好,但随着点数增加会被超越。
SAM的性能与不同因素(包括边界复杂性、强度差异等)相关。
对SAM进行针对特定医学任务的微调可以提高平均DICE性能,分别为ViT-B提高4.39%,ViT-H提高6.68%
Mazurowski M A, Dong H, Gu H, et al. Segment anything model for medical image analysis: an experimental study[J]. Medical Image Analysis, 2023, 89: 102918.【开源】
本文是直接实验验证,没有改变SAM,不过多解读,只关注结论
SAM的性能取决于数据集和任务的不同而有很大差异,对于某些医学影像数据集表现出色,而对于其他数据集则表现一般或较差。SAM在处理具有明确轮廓和较少模糊性提示的对象(如CT扫描中的器官分割)时表现更好。此外,SAM的性能在使用框提示时比使用点提示时更为突出。尽管SAM在单点提示设置中优于类似方法,但在提供多点提示进行迭代时,SAM的性能通常只有轻微提升,而其他方法的表现则有显著提高。研究还提供了SAM在所有测试数据集上的性能示例,包括迭代分割和在提示模糊情况下的行为。研究结论指出,SAM在特定医学影像数据集上展示了令人印象深刻的零样本分割性能,但在其他数据集上表现适中至较差,其在自动化医学影像分割中具有重要潜力,但在使用时需要谨慎。
Ma J, Wang B. Segment anything in medical images[J]. arXiv preprint arXiv:2304.12306, 2023.【开源】
本文是直接实验验证,没有改变SAM,不过多解读,只关注结论和局限性。
贡献:本文介绍了MedSAM(医学图像分割的基础模型),这是一种为医学图像分割设计的创新工具。MedSAM利用了超过一百万张医学图像构建的大规模数据集,能够在多种分割任务中显示出卓越的性能。与传统的专门化模型相比,MedSAM不仅展示了更好的通用性,而且在某些情况下甚至超越了这些模型。MedSAM的关键特点包括能够处理各种解剖结构、病理条件和医学成像方式,以及对用户提示(如边界框)的响应能力,从而实现精确的目标区域分割。
局限性:训练集中的模态不平衡,其中 CT、MRI和内窥镜图像在数据集中占主导地位。这可能会影响模型在较少代表性的模式(例如乳房X光检查)上的性能。另一个限制是它在分割血管状分支结构方面存在困难,因为在此设置中边界框提示可能不明确。
Wu J, Fu R, Fang H, et al. Medical sam adapter: Adapting segment anything model for medical image segmentation[J]. arXiv preprint arXiv:2304.12620, 2023.【开源】
本文同样不直接在医学数据集上微调,而是通过适配器改进。
【论文概述】
这篇论文的核心思想是开发和验证一种名为“医学SAM适配器”(Medical SAM Adapter, MSA)的新技术,用于提升医学图像分割的效能。作者们指出,虽然“分割任何事物模型”(Segment Anything Model, SAM)在图像分割领域表现出色,但在医学图像分割方面却表现不佳。为了解决这个问题,他们提出了一种简单但有效的适配技术,通过整合特定于医学领域的知识到分割模型中。MSA展示了在19种医学图像分割任务上的卓越性能,包括CT、MRI、超声波图像、眼底图像和皮肤镜图像等多种图像模态。这项工作不仅展示了使用参数高效的适配技术可以显著提高原始SAM模型的性能,而且还表明将强大的通用分割模型转移到医学应用领域是可行的。
【方法】
Zhang Y, Jiao R. Towards Segment Anything Model (SAM) for Medical Image Segmentation: A Survey[J]. arXiv preprint arXiv:2305.03678, 2023.【开源】
【文章概述】
尽管SAM在自然图像分割中表现出色,但由于医学图像与自然图像在结构复杂度、对比度和观察者间可变性方面的显著差异,SAM在医学图像分割中的适用性尚不清楚。这篇综述文章提供了对SAM在医学图像分割任务中应用的全面了解,包括其现有的性能、挑战、改进方向以及未来的发展潜力。
【论文中提到的将SAM应用的医学领域的几种方式】
提高对不同提示的鲁棒性
由于SAM直接应用于医学图像分割的性能不令人满意,许多研究集中于微调SAM的一小部分参数,如图像编码器、提示编码器和掩码解码器。这种微调的目的是提高SAM在特定医学图像分割任务上的可靠性和效果。
将 SAM 的可用性扩展到医学图像
医学图像通常有特定的格式,如NII和DICOM。为了简化SAM在这些格式上的使用,一些研究将SAM集成到常用的医学图像查看器中,例如3D Slicer,从而使研究人员能够在0.6秒的延迟内对医学图像进行分割。这种集成使得SAM能够自动地应用于连续的切片,从而提高其在医学图像处理中的实用性。
提高对不同提示的鲁棒性
尽管在医学数据集上微调SAM可以提高性能,但它仍然需要使用手动给出的框或点,这使得实现完全自动的医学图像分割变得困难。此外,最终的分割结果高度依赖于输入的提示,而且模型对错误的提示更为敏感。这表明需要进一步的方法来提高SAM对不同提示的鲁棒性。
利用SAM进行输入增强
由于SAM在直接应用于需要领域特定知识的医学图像分割任务时表现不佳,直接利用SAM生成的分割掩码来增强原始的医学图像输入。这种输入增强是通过融合功能实现的,目的是利用SAM生成的分割掩码来改善原始医学图像的分割性能。
【未来的发展】
总体来说,尽管SAM目前与领域特定模型相比性能不稳定,但它在建立医学图像分割基础模型方面展现了巨大的潜力,并有望作为一个高效且强大的工具,进一步协助临床应用。
【文章的彩蛋】
这篇文章的宝藏是其提供了一个不断更新的仓库,罗列了设计到的SAM应用于医学图像方面的论文和数据集。这里把仓库中的内容复制了一下,截至2023.12月:
Date | Authors | Title | Dataset |
---|---|---|---|
202311 | J. Ye et al. | SA-Med2D-20M Dataset: Segment Anything in 2D Medical Imaging with 20 Million masks (paper) | Link |
Date | Authors | Title | Code |
---|---|---|---|
202312 | W. Yue et al. | Part to Whole: Collaborative Prompting for Surgical Instrument Segmentation (paper) | Code |
202312 | ZM. Colbert et al. | Repurposing Traditional U-Net Predictions for Sparse SAM Prompting in Medical Image Segmentation (paper) | None |
202312 | W. Xie et al. | SAM Fewshot Finetuning for Anatomical Segmentation in Medical Images (paper) | None |
202312 | JG. Almeida et al. | Testing the Segment Anything Model on radiology data (paper) | None |
202312 | M. Barakat et al. | Towards SAMBA: Segment Anything Model for Brain Tumor Segmentation in Sub-Sharan African Populations (paper) | None |
202312 | Y. Zhang et al. | SQA-SAM: Segmentation Quality Assessment for Medical Images Utilizing the Segment Anything Model (paper) | Code |
202312 | S. Chen et al. | ASLseg: Adapting SAM in the Loop for Semi-supervised Liver Tumor Segmentation (paper) | None |
202312 | HE. Wong et al. | ScribblePrompt: Fast and Flexible Interactive Segmentation for Any Medical Image (paper) | Code |
202312 | Y. Zhang et al. | SemiSAM: Exploring SAM for Enhancing Semi-Supervised Medical Image Segmentation with Extremely Limited Annotations (paper) | None |
202312 | Y. Zhao et al. | Segment Anything Model-guided Collaborative Learning Network for Scribble-supervised Polyp Segmentation (paper) | None |
202311 | N. Li et al. | Segment Anything Model for Semi-Supervised Medical Image Segmentation via Selecting Reliable Pseudo-Labels (paper) | None |
202311 | X. Wei et al. | I-MedSAM: Implicit Medical Image Segmentation with Segment Anything (paper) | None |
202311 | Z. Shui et al. | Unleashing the Power of Prompt-driven Nucleus Instance Segmentation (paper) | Code |
202311 | M. Li and G. Yang et al. | Where to Begin? From Random to Foundation Model Instructed Initialization in Federated Learning for Medical Image Segmentation (paper) | None |
202311 | AK. Tyagi et al. | Guided Prompting in SAM for Weakly Supervised Cell Segmentation in Histopathological Images (paper) | Code |
202311 | Y. Du et al. | SegVol: Universal and Interactive Volumetric Medical Image Segmentation (paper) | Code |
202311 | DM. Nguyen et al. | On the Out of Distribution Robustness of Foundation Models in Medical Image Segmentation (paper) | None |
202311 | U. Israel et al. | A Foundation Model for Cell Segmentation (paper) | Code |
202311 | Q. Quan et al. | Slide-SAM: Medical SAM Meets Sliding Window (paper) | None |
202311 | Y. Zhang et al. | Segment Anything Model with Uncertainty Rectification for Auto-Prompting Medical Image Segmentation (paper) | Code |
202311 | Y. Wang et al. | SAMIHS: Adaptation of Segment Anything Model for Intracranial Hemorrhage Segmentation (paper) | Code |
202311 | H. Jiang et al. | GlanceSeg: Real-time microangioma lesion segmentation with gaze map-guided foundation model for early detection of diabetic retinopathy (paper) | None |
202311 | Y. Xu et al. | EviPrompt: A Training-Free Evidential Prompt Generation Method for Segment Anything Model in Medical Images (paper) | None |
202311 | DL. Ferreira and R. Arnaout | Are foundation models efficient for medical image segmentation? (paper) | Code |
202310 | H. Li et al. | Promise:Prompt-driven 3D Medical Image Segmentation Using Pretrained Image Foundation Models (paper) | Code |
202310 | D. Anand et al. | One-shot Localization and Segmentation of Medical Images with Foundation Models (paper) | None |
202310 | H. Wang et al. | SAM-Med3D (paper) | Code |
202310 | SK. Kim et al. | Evaluation and improvement of Segment Anything Model for interactive histopathology image segmentation (paper) | Code |
202310 | X. Chen et al. | SAM-OCTA: Prompting Segment-Anything for OCTA Image Segmentation (paper) | Code |
202310 | M. Peivandi et al. | Empirical Evaluation of the Segment Anything Model (SAM) for Brain Tumor Segmentation (paper) | None |
202310 | H. Ravishankar et al. | SonoSAM - Segment Anything on Ultrasound Images (paper) | None |
202310 | A. Ranem et al. | Exploring SAM Ablations for Enhancing Medical Segmentation in Radiology and Pathology (paper) | None |
202310 | S. Pandey et al. | Comprehensive Multimodal Segmentation in Medical Imaging: Combining YOLOv8 with SAM and HQ-SAM Models (paper) | None |
202309 | Y. Li et al. | nnSAM: Plug-and-play Segment Anything Model Improves nnUNet Performance (paper) | Code |
202309 | Y. Zhao et al. | MFS Enhanced SAM: Achieving Superior Performance in Bimodal Few-shot Segmentation (paper) | Code |
202309 | C. Wang et al. | SAM-OCTA: A Fine-Tuning Strategy for Applying Foundation Model to OCTA Image Segmentation Tasks (paper) | Code |
202309 | Y. Zhang et al. | 3D-U-SAM Network For Few-shot Tooth Segmentation in CBCT Images (paper) | None |
202309 | CJ. Chao et al. | Comparative Eminence: Foundation versus Domain-Specific Model for Cardiac Ultrasound Segmentation (paper) | None |
202309 | H. Ning et al. | An Accurate and Efficient Neural Network for OCTA Vessel Segmentation and a New Dataset (paper) | Code |
202309 | C. Chen et al. | MA-SAM: Modality-agnostic SAM Adaptation for 3D Medical Image Segmentation (paper) | Code |
202309 | P. Zhang and Y. Wang | Segment Anything Model for Brain Tumor Segmentation (paper) | None |
202309 | B. Fazekas et al. | Adapting Segment Anything Model (SAM) for Retinal OCT (paper) | None |
202309 | X. Lin et al. | SAMUS: Adapting Segment Anything Model for Clinically-Friendly and Generalizable Ultrasound Image Segmentation (paper) | Code |
202309 | X. Xing et al. | SegmentAnything helps microscopy images based automatic and quantitative organoid detection and analysis (paper) | Code |
202309 | NT. Bui et al. | SAM3D: Segment Anything Model in Volumetric Medical Images (paper) | Code |
202308 | Y. Zhang et al. | Self-Sampling Meta SAM: Enhancing Few-shot Medical Image Segmentation with Meta-Learning (paper) | None |
202308 | J. Cheng et al. | SAM-Med2D (paper) | Code |
202308 | C. Li et al. | Auto-Prompting SAM for Mobile Friendly 3D Medical Image Segmentation (paper) | None |
202308 | W. Feng et al. | Cheap Lunch for Medical Image Segmentation by Fine-tuning SAM on Few Exemplars (paper) | None |
202308 | Y. Zhang et al. | SamDSK: Combining Segment Anything Model with Domain-Specific Knowledge for Semi-Supervised Learning in Medical Image Segmentation (paper) | None |
202308 | A. Lou et al. | SAMSNeRF: Segment Anything Model (SAM) Guides Dynamic Surgical Scene Reconstruction by Neural Radiance Field (NeRF) (paper) | Code |
202308 | A. Archit et al. | Segment Anything for Microscopy (paper) | Code |
202308 | X. Yao et al. | False Negative/Positive Control for SAM on Noisy Medical Images (paper) | Code |
202308 | B. Fazekas et al. | SAMedOCT: Adapting Segment Anything Model (SAM) for Retinal OCT (paper) | None |
202308 | W. Yue et al. | SurgicalSAM: Efficient Class Promptable Surgical Instrument Segmentation (paper) | Code |
202308 | H. Zhang et al. | CARE: A Large Scale CT Image Dataset and Clinical Applicable Benchmark Model for Rectal Cancer Segmentation (paper) | Code |
202308 | Q. Wu et al. | Self-Prompting Large Vision Models for Few-Shot Medical Image Segmentation (paper) | Code |
202308 | A. Wang et al. | SAM Meets Robotic Surgery: An Empirical Study on Generalization, Robustness and Adaptation (paper) | None |
202308 | D. Shin et al. | CEmb-SAM: Segment Anything Model with Condition Embedding for Joint Learning from Heterogeneous Datasets (paper) | None |
202308 | R. Biswas | Polyp-SAM++: Can A Text Guided SAM Perform Better for Polyp Segmentation? (paper) | Code |
202308 | S. Cao et al. | TongueSAM: An Universal Tongue Segmentation Model Based on SAM with Zero-Shot (paper) | Code |
202308 | X. Li et al. | Leverage Weakly Annotation to Pixel-wise Annotation via Zero-shot Segment Anything Model for Molecular-empowered Learning (paper) | None |
202308 | JN. Paranjape et al. | AdaptiveSAM: Towards Efficient Tuning of SAM for Surgical Scene Segmentation (paper) | Code |
202308 | Z. Huang et al. | Push the Boundary of SAM: A Pseudo-label Correction Framework for Medical Segmentation (paper) | None |
202307 | J. Zhang et al. | SAM-Path: A Segment Anything Model for Semantic Segmentation in Digital Pathology (paper) | None |
202307 | MS. Hossain et al. | Robust HER2 Grading of Breast Cancer Patients using Zero-shot Segment Anything Model (SAM) (paper) | None |
202307 | C. Wang et al. | SAMMed : A medical image annotation framework based on large vision model (paper) | None |
202307 | G. Deng et al. | SAM-U: Multi-box prompts triggered uncertainty estimation for reliable SAM in medical image (paper) | None |
202307 | H. Kim et al. | Empirical Analysis of a Segmentation Foundation Model in Prostate Imaging (paper) | None |
202307 | X. Shi et al. | Cross-modality Attention Adapter: A Glioma Segmentation Fine-tuning Method for SAM Using Multimodal Brain MR Images (paper) | None |
202307 | C. Cui et al. | All-in-SAM: from Weak Annotation to Pixel-wise Nuclei Segmentation with Prompt-based Finetuning (paper) | None |
202306 | E. Kellener et al. | Utilizing Segment Anything Model for Assessing Localization of Grad-CAM in Medical Imaging (paper) | None |
202306 | F. H?rst et al. | CellViT: Vision Transformers for Precise Cell Segmentation and Classification (paper) | Code |
202306 | W. Lei et al. | MedLSAM: Localize and Segment Anything Model for 3D Medical Images (paper) | Code |
202306 | X. Hu et al. | How to Efficiently Adapt Large Segmentation Model (SAM) to Medical Images (paper) | Code |
202306 | S. Gong et al. | 3DSAM-adapter: Holistic Adaptation of SAM from 2D to 3D for Promptable Medical Image Segmentation (paper) | Code |
202306 | DMH. Nguyen et al. | LVM-Med: Learning Large-Scale Self-Supervised Vision Models for Medical Imaging via Second-order Graph Matching (paper) | Code |
202306 | S. Chai et al. | Ladder Fine-tuning approach for SAM integrating complementary network (paper) | Code |
202306 | L. Zhang et al. | Segment Anything Model (SAM) for Radiation Oncology (paper) | None |
202306 | G. Ning et al. | The potential of ‘Segment Anything’ (SAM) for universal intelligent ultrasound image guidance (paper) | None |
202306 | C. Shen et al. | Temporally-Extended Prompts Optimization for SAM in Interactive Medical Image Segmentation (paper) | None |
202306 | T. Shaharabany et al. | AutoSAM: Adapting SAM to Medical Images by Overloading the Prompt Encoder (paper) | None |
202306 | Y. Gao et al. | DeSAM: Decoupling Segment Anything Model for Generalizable Medical Image Segmentation (paper) | Code |
202305 | D. Lee et al. | IAMSAM : Image-based Analysis of Molecular signatures using the Segment-Anything Model (paper) | Code |
202305 | M. Hu et al. | BreastSAM: A Study of Segment Anything Model for Breast Tumor Detection in Ultrasound Images (paper) | None |
202305 | J. Wu | PromptUNet: Toward Interactive Medical Image Segmentation (paper) | Code |
202305 | Y. Li et al. | Polyp-SAM: Transfer SAM for Polyp Segmentation (paper) | Code |
202305 | C. Mattjie et al. | Exploring the Zero-Shot Capabilities of the Segment Anything Model (SAM) in 2D Medical Imaging: A Comprehensive Evaluation and Practical Guideline (paper) | None |
202305 | D. Cheng et al. | SAM on Medical Images: A Comprehensive Study on Three Prompt Modes (paper) | None |
202304 | A. Wang et al. | SAM Meets Robotic Surgery: An Empirical Study in Robustness Perspective (paper) | None |
202304 | Y. Huang et al. | Segment Anything Model for Medical Images? (paper) | None |
202304 | M. Hu et al. | SkinSAM: Empowering Skin Cancer Segmentation with Segment Anything Model (paper) | None |
202304 | B. Wang et al. | GazeSAM: What You See is What You Segment (paper) | Code |
202304 | K. Zhang and D. Liu | Customized Segment Anything Model for Medical Image Segmentation (paper) | Code |
202304 | Z. Qiu et al. | Learnable Ophthalmology SAM (paper) | Code |
202304 | P. Shi et al. | Generalist Vision Foundation Models for Medical Imaging: A Case Study of Segment Anything Model on Zero-Shot Medical Segmentation (paper) | None |
202304 | J. Wu et al. | Medical SAM Adapter: Adapting Segment Anything Model for Medical Image Segmentation (paper) | Code |
202304 | J. Ma and B. Wang | Segment Anything in Medical Images (paper) | Code |
202304 | Y. Zhang et al. | Input Augmentation with SAM: Boosting Medical Image Segmentation with Segmentation Foundation Model (paper) | None |
202304 | MA. Mazurowski et al. | Segment Anything Model for Medical Image Analysis: an Experimental Study (paper) | Code |
202304 | S. He et al. | Accuracy of Segment-Anything Model (SAM) in medical image segmentation tasks (paper) | None |
202304 | T. Chen et al. | SAM Fails to Segment Anything? – SAM-Adapter: Adapting SAM in Underperformed Scenes: Camouflage, Shadow, Medical Image Segmentation, and More (paper) | Code |
202304 | C. Hu and X. Li | When SAM Meets Medical Images: An Investigation of Segment Anything Model (SAM) on Multi-phase Liver Tumor Segmentation (paper) | None |
202304 | F. Putz et al. | The “Segment Anything” foundation model achieves favorable brain tumor autosegmentation accuracy on MRI to support radiotherapy treatment planning (paper) | None |
202304 | T. Zhou et al. | Can SAM Segment Polyps? (paper) | Code |
202304 | Y. Liu et al. | SAMM (Segment Any Medical Model): A 3D Slicer Integration to SAM (paper) | Code |
202304 | S. Roy et al. | SAM.MD: Zero-shot medical image segmentation capabilities of the Segment Anything Model (paper) | None |
202304 | S. Mohapatra et al. | SAM vs BET: A Comparative Study for Brain Extraction and Segmentation of Magnetic Resonance Images using Deep Learning (paper) | None |
202304 | R. Deng et al. | Segment Anything Model (SAM) for Digital Pathology: Assess Zero-shot Segmentation on Whole Slide Imaging (paper) | None |
Shaharabany T, Dahan A, Giryes R, et al. AutoSAM: Adapting SAM to Medical Images by Overloading the Prompt Encoder[J]. arXiv preprint arXiv:2306.06370, 2023.
【文章概述】
文章介绍了一种名为AutoSAM的方法,旨在改善Segment Anything Model(SAM)在医学影像领域的应用。SAM虽然在图像分割方面有出色的表现,但在处理非自然图像(如医学影像)时效果不佳。AutoSAM的核心思想是替换SAM的条件编码器(prompt encoder)为一个新的prompts生成编码器,这个编码器直接处理输入图像,而不需要进一步微调SAM模型。
图2展示了AutoSAM的框架。在这个框架中,Segment Anything Model (SAM) 的提示编码器(prompt encoder)被替换成了自定义的编码器,而图像编码器(image encoder)和掩码解码器(mask decoder)则保持不变。这个图解说明了如何通过替换SAM的一个组成部分来适应医学图像的处理,而不需要改变其他核心部件。
【方法】
AutoSAM采用了一个辅助的提示编码网络(auxiliary prompt encoder network),通过输入图像生成SAM的替代提示(surrogate prompt)。
【实验】
结果
【结论】