Abstract:Recent advancements in computational pathology and artificial intelligence have significantly improved whole slide image (WSI) classification. However, the gigapixel resolution of WSIs and the scarcity of manual annotations present substantial challenges. Multiple instance learning (MIL) is a promising weakly supervised learning approach for WSI classification. Recently research revealed employing pseudo bag augmentation can encourage models to learn various data, thus bolstering models' performance. While directly inheriting the parents' labels can introduce more noise by mislabeling in training. To address this issue, we translate the WSI classification task from weakly supervised learning to semi-weakly supervised learning, termed SWS-MIL, where adaptive pseudo bag augmentation (AdaPse) is employed to assign labeled and unlabeled data based on a threshold strategy. Using the "student-teacher" pattern, we introduce a feature augmentation technique, MergeUp, which merges bags with low-priority bags to enhance inter-category information, increasing training data diversity. Experimental results on the CAMELYON-16, BRACS, and TCGA-LUNG datasets demonstrate the superiority of our method over existing state-of-the-art approaches, affirming its efficacy in WSI classification.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in the classification of whole - slide images (WSI), due to the high resolution of WSI and the scarcity of manual annotations, it is difficult to effectively apply supervised learning methods. Specifically, the paper proposes an improved weakly - supervised learning framework - semi - weakly - supervised learning (SWSL), aiming to improve the performance of WSI classification through adaptive pseudo - package assignment (AdaPse) and feature enhancement techniques (MergeUp). ### Main problems 1. **High resolution and memory limitations**: WSI has an extremely high resolution, which makes them face huge memory limitations in processing and is difficult to be directly processed like natural images. 2. **Scarcity of manual annotations**: Manual annotations of WSI are very scarce, which limits the application of fully - supervised learning methods. 3. **Label noise in pseudo - package assignment**: Existing pseudo - package assignment methods introduce a large amount of label noise while increasing data diversity, affecting the training effect of the model. ### Solutions 1. **Adaptive pseudo - package assignment (AdaPse)**: - Through an adaptive threshold strategy, filter out high - confidence pseudo - packages and discard those pseudo - packages whose predictions are inconsistent with the parent - package labels. - This method transforms weakly - supervised learning tasks into semi - weakly - supervised learning tasks, reducing the impact of label noise. 2. **Feature enhancement techniques (MergeUp)**: - Merge packages of different categories while retaining high - priority labels, enabling the model to learn the relationships between different categories. - This method is especially suitable for non - mutually - exclusive tasks and significantly increases the diversity of training data. ### Experimental results The paper conducted experiments on three datasets, CAMELYON - 16, BRACS, and TCGA - LUNG. The results show that the proposed method outperforms existing state - of - the - art methods in multiple indicators, especially in AUC and ACC indicators. ### Conclusions The method proposed in the paper effectively solves the key problems in WSI classification through adaptive pseudo - package assignment and feature enhancement techniques, improving the performance and robustness of the model. Future work will further compress the module size, improve the training speed, and explore feature fusion methods for mutually - exclusive datasets.

MergeUp-augmented Semi-Weakly Supervised Learning for WSI Classification

Weakly Supervised Instance Segmentation Using Multi-Prior Fusion.

Learning Hybrid Negative Probability Model for Weakly-Supervised Whole Slide Image Recognition.

A universal multiple instance learning framework for whole slide image analysis

Pseudo-Bag Mixup Augmentation for Multiple Instance Learning-Based Whole Slide Image Classification

Iterative multiple instance learning for weakly annotated whole slide image classification

The Whole Pathological Slide Classification via Weakly Supervised Learning

Enhancing Weakly Supervised Semantic Segmentation with Multi-label Contrastive Learning and LLM Features Guidance

Semantic-Similarity Collaborative Knowledge Distillation Framework for Whole Slide Image Classification

ReMix: A General and Efficient Framework for Multiple Instance Learning Based Whole Slide Image Classification

Enhanced Pooling for Weakly Supervised Gigapixel WSI Training Improves Classification and Lesion Localization

Weakly Supervised Breast Cancer Classification on WSI Using Transformer and Graph Attention Network

Bayesian Collaborative Learning for Whole-Slide Image Classification

Self-Supervised Representation Distribution Learning for Reliable Data Augmentation in Histopathology WSI Classification

AdvMIL: Adversarial multiple instance learning for the survival analysis on whole-slide images

Enhancing Whole Slide Image Classification with Discriminative and Contrastive Learning

TDT-MIL: a framework with a dual-channel spatial positional encoder for weakly-supervised whole slide image classification

Task-specific Fine-tuning Via Variational Information Bottleneck for Weakly-supervised Pathology Whole Slide Image Classification

Iteratively Coupled Multiple Instance Learning from Instance to Bag Classifier for Whole Slide Image Classification

Gigapixel Whole-Slide Images Classification using Locally Supervised Learning

Cervical Cytology Classification with Coarse Labels Based on Two-Stage Weakly Supervised Contrastive Learning Framework.