Abstract:Deep neural networks have reached remarkable achievements in medical image processing tasks, specifically in classifying and detecting various diseases. However, when confronted with limited data, these networks face a critical vulnerability, often succumbing to overfitting by excessively memorizing the limited information available. This work addresses the challenge mentioned above by improving the supervised contrastive learning method leveraging both image-level labels and domain-specific augmentations to enhance model robustness. This approach integrates self-supervised pre-training with a two-stage supervised contrastive learning strategy. In the first stage, we employ a modified supervised contrastive loss that not only focuses on reducing false negatives but also introduces an elimination effect to address false positives. In the second stage, a relaxing mechanism is introduced that refines positive and negative pairs based on similarity, ensuring that only relevant image representations are aligned. We evaluate our method on the BreakHis dataset, which consists of breast cancer histopathology images, and demonstrate an increase in classification accuracy by 1.45% in the image level, compared to the state-of-the-art method. This improvement corresponds to 93.63% absolute accuracy, highlighting the effectiveness of our approach in leveraging properties of data to learn more appropriate representation space.

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve This paper aims to address several key issues in the classification of breast cancer histopathological images: 1. **Data Scarcity Issue**: In the medical field, labeled data is limited, posing challenges for supervised learning methods. To overcome this issue, the researchers draw on self-supervised learning techniques and utilize the similarity of labeled data and images in the representation space to improve the model. 2. **Overfitting Issue**: When faced with limited data, deep neural networks are prone to overfitting. This paper improves the supervised contrastive learning method by combining image-level labels and domain-specific data augmentation techniques to enhance the model's robustness. 3. **Generalization Ability Issue**: To improve the model's generalization ability, the researchers employ various data augmentation techniques specific to histopathological datasets and HE staining. ### Main Contributions - Proposed a two-stage supervised contrastive learning strategy. The first stage improves the supervised contrastive loss function by not only reducing false negatives but also introducing a mechanism to eliminate false positives. The second stage introduces a relaxation mechanism to adjust positive and negative sample pairs based on similarity. - Conducted experimental evaluation on the BreakHis dataset, achieving a 1.45% improvement in image-level classification accuracy, reaching an absolute accuracy of 93.63%. - The study demonstrates that the proposed method can effectively utilize data attributes to learn a more suitable representation space. ### Method Overview - Utilized SimCLR technology to pre-train the model and used these pre-trained weights as the initial weights for the representation learning stage. - The representation learning stage includes two supervised contrastive learning stages, ultimately completing the classification task through supervised fine-tuning. - Introduced various data augmentation techniques such as random cropping, color jittering, and Gaussian blur to enhance the model's generalization ability. - In the relaxation stage, adjusted positive and negative sample pairs based on similarity in the representation space to further optimize the model. - The fine-tuning stage includes a primary classification task and an auxiliary task to improve the model's robustness to different HE staining. ### Experimental Results - Achieved superior image-level and patient-level accuracy on the BreakHis dataset compared to existing methods, with improvements of 1.45% and 0.31%, respectively. - Validated the model's generalization ability using the BACH dataset, achieving over 90% accuracy even with limited training.

Classification of Breast Cancer Histopathology Images using a Modified Supervised Contrastive Learning Method

Enhancing breast cancer classification via histopathological image analysis: Leveraging self-supervised contrastive learning and transfer learning

Breast cancer pathological image classification based on deep learning

Optimizing the Performance of Breast Cancer Classification by Employing the Same Domain Transfer Learning from Hybrid Deep Convolutional Neural Network Model

Improved Breast Cancer Classification through Combining Transfer Learning and Attention Mechanism

Breast tumor classification based on self-supervised contrastive learning from ultrasound videos

Enhanced Dual Contrast Representation Learning with Cell Separation and Merging for Breast Cancer Diagnosis

Breast Tumor Classification Using EfficientNet Deep Learning Model

Compatible-domain Transfer Learning for Breast Cancer Classification with Limited Annotated Data

Breast Cancer Histopathology Images Classification Through Multi-View Augmented Contrastive Learning and Pre-Learning Knowledge Distillation

Breast Cancer Histopathological Image Classification with Adversarial Image Synthesis

A New Transfer Learning Based Approach to Magnification Dependent and Independent Classification of Breast Cancer in Histopathological Images

Imbalanced Histopathological Breast Cancer Image Classification with Convolutional Neural Network

Classification of Breast Cancer Lesions in Ultrasound Images by using Attention Layer and loss Ensembles in Deep Convolutional Neural Networks

Classification of breast cancer histopathological image with deep residual learning

Attention-Based Deep Learning Approach for Breast Cancer Histopathological Image Multi-Classification

Classification of Breast Cancer Lesions in Ultrasound Images by Using Attention Layer and Loss Ensemble in Deep Convolutional Neural Networks

Breast Cancer Histopathological Image Classification Using a Hybrid Deep Neural Network.

Supervised Contrastive Vision Transformer for Breast Histopathological Image Classification

Breast lesions classifications of mammographic images using a deep convolutional neural network-based approach

Classifying breast cancer using transfer learning models based on histopathological images