深度学习和骨关节炎特征分类基础模型嵌入的比较评估
Mohammadreza Chavoshi1, Hari Trivedi1, Janice Newsome1
1Department of Radiology, Emory University, Atlanta, GA, USA.
Journal of imaging informatics in medicine
|September 2, 2025
概括
监督深度学习模型比基础模型更好地分类膝关节关节炎的放射性特征. 在Kellgren-Lawrence分级中,ConvNeXt获得了最高的性能,在肌肉骨成像分析中表现出卓越的准确性.
科学领域:
- 医学成像
- 人工智能
- 整形医学
背景情况:
- 基础模型 (FM) 为监督深度学习 (DL) 提供灵活,可通用的替代方案,不需要大型标记数据集.
- 对于膝关节骨关节炎的放射性特征分类来说,研究FM嵌入与监督的DL是非常重要的.
研究的目的:
- 为了比较受监督的DL模型和预训练的FM嵌入器在分类膝关节关节炎放射性特征的性能.
- 评估不同模型的有效性,包括ResNet18,ConvNeXt-Small,BiomedCLIP和RAD-DINO,用于此分类任务.
主要方法:
- 从骨关节炎倡议数据集中分析了44,985张膝盖X射线图.
- 训练卷积神经网络 (ResNet18,ConvNeXt-Small) 并使用XGBoost分类器的FM嵌入式 (BiomedCLIP,RAD-DINO).
- 使用二进制和多类分类指标进行全面评估,包括科恩的kappa和AUC,以及用于可视化的Grad-CAM.
主要成果:
- 在所有分类任务中,监督的DL模型显著优于基于FM的方法.
- 在预测凯尔格伦-劳伦斯等级 (加权的科恩卡帕为0.880) 和二进制任务 (更高的AUC) 中,ConvNeXt-Small取得了最高的表现.
- FM模型 (BiomedCLIP,RAD-DINO) 显示了类似的性能;BiomedCLIP的零射击分类实现了91.14%的准确性,与图像质量相关的故障.
结论:
- 监督DL模型,特别是ConvNeXt,在肌肉骨成像中表现出优异的细粒度放射特征分类性能.
- 基础模型对辅助成像任务有希望,但当预训练表现有限时,对于特定特征的分类,它们目前的效率低于监督DL.
- DL模型可靠地关注临床相关区域,Grad-CAM可视化证实了其可解释性和临床实用性.
相关概念视频
Functional Classification of Joints
4.6K
Functional Classification of Joints
The functional classification of joints is determined by the amount of mobility between the adjacent bones. Joints are functionally classified as a synarthrosis or immobile joint, an amphiarthrosis or slightly moveable joint, or as a diarthrosis, a freely moveable joint. Fibrous and cartilaginous joints can be functionally classified as either synarthroses or amphiarthroses, whereas all synovial joints are classified as diarthroses.
Synarthrosis
An...
The functional classification of joints is determined by the amount of mobility between the adjacent bones. Joints are functionally classified as a synarthrosis or immobile joint, an amphiarthrosis or slightly moveable joint, or as a diarthrosis, a freely moveable joint. Fibrous and cartilaginous joints can be functionally classified as either synarthroses or amphiarthroses, whereas all synovial joints are classified as diarthroses.
Synarthrosis
An...
4.6K
Classification of Bones
7.0K
The bones of the human skeletal system are of varied shapes, sizes, and functions. They can be classified based on their shape and function into four major classes: long bones, short bones, flat bones, and irregular bones. Some classifications include a fifth type, the sesamoid bones, as a separate class, whereas others categorize them under short bones.
Long and Short Bones
The appendicular skeleton, particularly the upper and lower limbs, is primarily made of long and short bones. The...
Long and Short Bones
The appendicular skeleton, particularly the upper and lower limbs, is primarily made of long and short bones. The...
7.0K
Knee Joint
2.2K
The knee joint is the most complicated joint in the body. It consists of three articulations– two tibiofemoral and one patellofemoral. As is characteristic of synovial joints, the knee joint has a thin articular capsule that partially surrounds this joint cavity. Additionally, several ligaments, muscles, and cartilaginous structures support the movement of the knee.
A total of seven ligaments support the knee joint. The patellar ligament, which is also attached to the quadriceps femoris...
A total of seven ligaments support the knee joint. The patellar ligament, which is also attached to the quadriceps femoris...
2.2K
Structural Classification of Joints
4.1K
Joints, also known as articulations, are classified based on their structural characteristics, i.e., based on whether the articulating surfaces of the adjacent bones are directly connected by fibrous connective tissue or cartilage, or whether the articulating surfaces contact each other within a fluid-filled joint cavity. These differences serve to divide the joints of the body into three structural classifications.
A fibrous joint is where the adjacent bones are united by fibrous connective...
A fibrous joint is where the adjacent bones are united by fibrous connective...
4.1K


