GS-MGF-LPLs:计算机视觉辅助的不确定性甲状腺结节超声图像分割方法
DOI:
https://doi.org/10.70693/cjst.v1i3.1324Abstract
不确定性甲状腺结节（indeterminate thyroid nodules，ITNs）的检测对患者生存预后至关重要。超声检查作为其诊断评估的金标准，在实际应用中面临医学图像细粒度分割资源消耗大、标注数据稀缺的挑战。本研究创新性地提出网格搜索聚类驱动的多门控局部块学习模型（简称GS-MGF-LPLs）。该模型通过重构双阈值快速搜索策略的块深度全卷积网络架构，将输入空间划分为潜在簇和噪声簇两类特征空间，并基于输出空间的局部簇构建目标模型。其技术突破体现在三个层面：（1）采用门控融合模型实现扫描参数的输入层前自动分割，以恒定连接权重表征不同类别簇的预测权重；（2）在加权层后集成多网格搜索局部块学习系统，通过多系统投票机制生成决策；（3）输出层通过解码分割结果获取块预测输出，运用动态加权平均算法保持特征多样性。基于专门构建的网格搜索耦合分割数据集验证表明，本模型在四类ITNs分割任务中取得突破性性能：准确率达99.54%，精确率为86.06%，F值与Dice系数均达到84.26%，显著优于当前五种主流分割方法（p<0.01）。该成果不仅为ITNs的早期精准诊断提供了可靠的技术支撑，其创新的特征空间解耦思想和多系统集成架构更为医学图像分析领域提供了方法论参考。目前，该模型已进入多中心临床验证阶段，有望成为辅助医生诊断决策的有力工具。
References
Pani K, Chawla I. A hybrid approach for multi modal brain tumor segmentation using two phase transfer learning, SSL and a hybrid 3DUNET[J]. Computers and Electrical Engineering, 2024, 118(PB).
李志强, 田少波, 王 伟等. 基于生成对抗网络的图像识别与增强技术研究[J]. 科技与创新, 2025(15): 143-146.
Pani K, Chawla I. A hybrid approach for multi modal brain tumor segmentation using two phase transfer learning, SSL and a hybrid 3DUNET[J]. Computers and Electrical Engineering, 2024, 118(PB).
Ning Z, Mao Y, Xu X, et al. DC-Net: Decomposing and coupling saliency map for lesion segmentation in ultrasound images[J]. Engineering Applications of Artificial Intelligence, 2025, 148.
王 楠, 林绍辉, 齐福霖等. 基于自监督学习的医学影像异常检测[J]. 计算机辅助设计与图形学学报, 2025, 37(03): 474-483.
龚凯琳, 张利丽, 宋佳佳等. 基于人工智能斑块分割超声图像的影像组学在颈动脉斑块稳定性评估中的应用[J]. 临床神经外科杂志, 2021, 18(01): 1-4.
Leng Z, Jia W, Chen B, et al. Multi-modal feature fusion: A hybrid framework for lung cancer subtype classification using CT imaging with radiomic and deep features[J]. Journal of Radiation Research and Applied Sciences, 2025, 18(3).
Zhang Y J, Li G D, Wu L L, et al. A lightweight self-supervised learning segmentation model for variable and complex high-resolution remote sensing images[J]. Applied Soft Computing, 2024, 165.
Yin L, Liu R. A normalized deep neural network with self-attention mechanisms based multi-objective multi-verse optimization algorithm for economic dispatch[J]. Applied Energy, 2025, 383.
Wang X, Li W, He X. MTDiff: Visual anomaly detection with multi-scale diffusion models[J]. Knowledge-Based Systems, 2024, 302.
Zhong L Y, Lu J L, Chen Z L, et al. Adaptive multi-channel contrastive graph convolutional network with graph and feature fusion[J]. Information Sciences, 2024, 658.
Yan Z, Ye Z, Ge J, et al. DocExtractNet: A novel framework for enhanced information extraction from business documents[J]. Information Processing and Management, 2025, 62(3).
周 涛, 党 培, 陆惠玲等. 跨模态跨尺度跨维度的PET/CT图像的Transformer分割模型[J]. 电子与信息学报, 2023, 45(10): 3529-3537.
左宪禹, 田展硕, 殷梦晗等. 基于残差扩散模型的遥感超分辨率图像生成研究[J]. 河南师范大学学报(自然科学版), 2025, 53(03): 58-65+178.
刘 飞, 张久楼, 金若帆等. 基于深度学习方法建立颞下颌关节核磁共振影像的自动分割模型[J]. 口腔医学, 2025, 45(06): 445-452.
Forien R, Pang G, Pardoux É. Multi-patch multi-group epidemic model with varying infectivity[J]. Probability, Uncertainty and Quantitative Risk, 2022, 7(04): 333-364.
魏盼丽, 王红斌. 融合关键信息与专家网络的生成式文本摘要[J]. 吉林大学学报(理学版), 2024, 62(04): 951-959.
槐文信, 耿 川, 曾玉红等. Analytical solutions for transverse distributions of stream-wise velocity in turbulent flow in rectangular channel with partial vegetation[J]. Applied Mathematics and Mechanics(English Edition), 2011, 32(04): 459-468.
Liu H, Ji H, Zhang J, et al. Identification algorithm of low-count energy spectra under short-duration measurement based on heterogeneous sample transfer[J]. Nuclear Science and Techniques, 2025, 36(03): 63-77.
姜永军, 李红玲, 阮 萍等. 基于深度学习的多亚型腹膜后软组织肉瘤诊断[J]. 中山大学学报(自然科学版中英文), 2025, 64(03): 156-164.
Ou J, Zhang J, Alswadeh M, et al. Advancing osteoarthritis research: the role of AI in clinical, imaging and omics fields[J]. Bone Research, 2025, 13(02): 307-319.
崔文成, 杨 丹, 邵 虹. 基于双路特征的宫颈细胞核分割[J]. 电子测量技术, 2023, 46(06): 129-136.
Tang J, Chen X, Fan L, et al. LN-DETR: An efficient Transformer architecture for lung nodule detection with multi-scale feature fusion[J]. Neurocomputing, 2025, 633.
Zhang B, Li M, Yu L, et al. Diffusion tensor imaging of spinal microstructure in healthy adults: improved resolution with the readout segmentation of long variable echo-trains[J]. Neural Regeneration Research, 2017, 12(12): 2067-2070.
黄鑫涛, 曹 力, 蔡有城等. 联合多尺度与注意力模型的双分支局部特征提取方法[J]. 小型微型计算机系统, 2023, 44(06): 1297-1303.
XU H, CHEN Z, WU Y, et al. Noninvasive high-resolution deep-brain photoacoustic imaging with a negatively focused fiber-laser ultrasound transducer[J]. Photonics Research, 2024, 12(12): 2996-3005.
陈 芳, 赵 喆, 张道强等. 特征重加权U-Net超声图像分割的骨结构重建与增强现实显示[J]. 计算机辅助设计与图形学学报, 2021, 33(03): 448-456.
陈 昊, 郭文普, 康 凯. 通道门控Res2Net卷积神经网络自动调制识别[J]. 电讯技术, 2023, 63(12): 1869-1875.
Star S P S C, Inbamalar T, Milton A. Automatic semantic segmentation of breast cancer in DCE-MRI using DeepLabV3+ with modified ResNet50[J]. Biomedical Signal Processing and Control, 2025, 99.
张冀豫, 许家佗, 屠立平等. 基于自适应权重多模态中西医数据融合方法的冠心病血管阻塞程度预测模型的构建与评价（英文）[J]. Digital Chinese Medicine, 2025, 8(02): 163-173.
毕 玉, 彭 梅, 方明娣. TI-RADS联合S-Detect技术对甲状腺结节诊断价值的探讨[J]. 中国超声医学杂志, 2021, 37(04): 364-366.
Yamazaki M, Kasagi A, Tabuchi A, et al. Yet Another Accelerated SGD: ResNet-50 Training on ImageNet in 74.7 seconds[J]. CoRR, 2019, abs/1903.12650.
吴志泽, 陈 鑫, 徐 童等. 基于动态多粒度图卷积网络的人体骨架行为识别[J]. 中国图象图形学报, 2025, 30(08): 2822-2834.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2025 杨 杰, 刘贤贤, 田洪瑜, 何志容, 康 慷, 陈 宏
This work is licensed under a Creative Commons Attribution 4.0 International License.
