《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (9): 3005-3016.DOI: 10.11772/j.issn.1001-9081.2025080960

• 多媒体计算与计算机仿真 • 上一篇    

双向协同蒸馏多实例学习的肺癌图像分类方法

李锦昌1, 贾伟1,2(), 赵雪芬1,2   

  1. 1.宁夏大学 信息工程学院,银川 750021
    2.宁夏“东数西算”人工智能与信息安全重点实验室(宁夏大学),银川 750021
  • 收稿日期:2025-08-22 修回日期:2025-12-17 接受日期:2026-01-04 发布日期:2026-02-12 出版日期:2026-09-10
  • 通讯作者: 贾伟
  • 作者简介:李锦昌(2000—),男,江西宜春人,硕士研究生,主要研究方向:医学图像处理与分析、计算病理学
    贾伟(1980—),男,宁夏银川人,副教授,博士,CCF会员,主要研究方向:医学图像处理与分析、计算病理学
    赵雪芬(1983—),女,宁夏隆德人,副教授,博士,主要研究方向:医学图像处理与分析。
  • 基金资助:
    宁夏自然科学基金资助项目(2024AAC03063)

Lung cancer image classification method via bidirectional collaborative distillation and multi-instance learning

Jinchang LI1, Wei JIA1,2(), Xuefen ZHAO1,2   

  1. 1.School of Information Engineering,Ningxia University,Yinchuan Ningxia 750021,China
    2.Ningxia Key Laboratory of Artificial Intelligence and Information Security for Channeling Computing Resources from the East to the West (Ningxia University),Yinchuan Ningxia 750021,China
  • Received:2025-08-22 Revised:2025-12-17 Accepted:2026-01-04 Online:2026-02-12 Published:2026-09-10
  • Contact: Wei JIA
  • About author:LI Jinchang, born in 2000, M. S. candidate. His research interests include medical image processing and analysis, computational pathology.
    JIA Wei, born in 1980, Ph. D., associate professor. His research interests include medical image processing and analysis, computational pathology.
    ZHAO Xuefen, born in 1983, Ph. D., associate professor. Her research interests include medical image processing and analysis.
  • Supported by:
    Natural Science Foundation of Ningxia(2024AAC03063)

摘要:

针对现有的肺癌全切片图像(WSI)分类中,基于知识蒸馏架构的多实例学习(MIL)方法难以有效处理混杂的难分类和易分类实例,导致模型学习能力不足与性能不均衡的问题,结合多任务学习思想,提出一种双向协同蒸馏多实例学习的肺癌全切片图像分类方法(BCD-MIL)。首先,设计一种包括多学生模型与教师模型的双向协同蒸馏框架(BCDF),以满足肺癌病理图像中易分类与难分类病理实例的差异化学习需求,通过多个学生模型分担不同任务,增强模型的学习能力并解决性能不均衡的问题,同时,通过与教师模型的良性循环提升模型的分类性能;其次,设计一种多角度实例挖掘的并行结构(MA-IMP),以匹配肺癌病理特征在细胞形态和组织纹理等维度的异质性差异,通过从多个不同角度进行实例挖掘,避免单一角度造成的实例挖掘偏差;最后,提出一种动态阶段感知的指数移动平均蒸馏(DSA-EMA)算法,根据肺癌病理图像大规模实例训练的阶段特性,通过在训练阶段动态调整蒸馏参数,优化教师模型权重更新,提高训练效率与模型性能。实验结果表明,与同样基于知识蒸馏和实例挖掘的多实例学习方法的MHIM-MIL(MIL framework with Masked Hard Instance Mining)相比,BCD-MIL在TCGA(The Cancer Genome Atlas)和CPTAC(Clinical Proteomic Tumor Analysis Consortium)数据集上的曲线下面积(AUC)、准确率、F1-Score和召回率分别提升了1.92、2.37、3.43和1.75个百分点,以及1.11、6.29、6.94和12.85个百分点,验证了该方法的有效性;同时,轻量化的蒸馏架构降低了模型的参数规模与推理耗时,从而在保证性能提升的前提下实现了高效部署,为肺癌WSI分类提供了可靠依据。

关键词: 肺癌, 全切片图像分类, 多实例学习, 知识蒸馏, 实例挖掘

Abstract:

To address the issue in the existing Whole Slide Image (WSI) classification for lung cancer, where Multi-Instance Learning (MIL) methods based on knowledge distillation architecture struggle to handle the mixture of hard and easy instances effectively, leading to insufficient model learning capability and imbalanced performance, a lung cancer whole slide image classification method based on Bidirectional Collaborative Distillation and Multi-Instance Learning (BCD-MIL) was proposed by incorporating the concept of multi-task learning. First, a Bidirectional Collaborative Distillation Framework (BCDF) comprising multiple student models and a teacher model was designed, so as to satisfy the differentiated learning requirements for easy and hard pathological instances in lung cancer histopathological images, and by distributing different tasks to multiple student models, the model's learning capacity was enhanced while solving performance imbalance problem. Concurrently, classification performance was improved through a virtuous cycle with the teacher model. Second, a Multi-Angle Instance Mining Parallel architecture (MA-IMP) was designed to match the heterogeneous differences in lung cancer pathological features across dimensions such as cell morphology and tissue texture, and by conducting instance mining from multiple perspectives, the mining bias caused by a single perspective was avoided. Finally, a Dynamic Stage-Aware Exponential Moving Average distillation (DSA-EMA) algorithm was proposed to optimize weight update of the teacher model, and improve training efficiency and model performance based on the stage characteristics of large-scale instance training of lung cancer histopathological images through adjusting distillation parameters in the training phase dynamically. Experimental results show that compared to the MIL framework with Masked Hard Instance Mining (MHIM-MIL) method, which is also based on knowledge distillation and instance mining, BCD-MIL achieves improvements of 1.92, 2.37, 3.43, and 1.75 percentage points in Area Under Curve (AUC), accuracy, F1-Score, and recall, respectively, on The Cancer Genome Atlas (TCGA) dataset, and improvements of 1.11, 6.29, 6.94, and 12.85 percentage points in four metrics, respectively, on the Clinical Proteomic Tumor Analysis Consortium (CPTAC) dataset; validating the effectiveness of the proposed method. Furthermore, the lightweight distillation architecture reduces model parameter size and inference time, thereby enabling efficient deployment while ensuring performance gains, and providing a reliable basis for lung cancer WSI classification.

Key words: lung cancer, Whole Slide Image (WSI) classification, Multi-Instance Learning (MIL), knowledge distillation, instance mining

中图分类号: