BACH: Grand Challenge on Breast Cancer Histology Images

Abstract: Breast cancer is the most common invasive cancer in women, affecting more than 10% of women worldwide. Microscopic analysis of a biopsy remains one of the most important methods to diagnose the type of breast cancer. This requires specialized analysis by pathologists, in a task that i) is highly time- and cost-consuming and ii) often leads to nonconsensual results. The relevance and potential of automatic classification algorithms using hematoxylin-eosin stained histopathological images has already been demonstrated, but the reported results are still sub-optimal for clinical use. With the goal of advancing the state-of-the-art in automatic classification, the Grand Challenge on BreAst Cancer Histology images (BACH) was organized in conjunction with the 15th International Conference on Image Analysis and Recognition (ICIAR 2018). A large annotated dataset, composed of both microscopy and whole-slide images, was specifically compiled and made publicly available for the BACH challenge. Following a positive response from the scientific community, a total of 64 submissions, out of 677 registrations, effectively entered the competition. From the submitted algorithms it was possible to push forward the state-of-the-art in terms of accuracy (87%) in automatic classification of breast cancer with histopathological images. Convolutional neuronal networks were the most successful methodology in the BACH challenge. Detailed analysis of the collective results allowed the identification of remaining challenges in the field and recommendations for future developments. The BACH dataset remains publically available as to promote further improvements to the field of automatic classification in digital pathology.

• The paper introduces a competition evaluating CNN models for classifying microscopy images into four breast cancer categories with top accuracy of around 87%.
• It employs transfer learning with architectures like VGG, Inception, ResNet, and DenseNet to achieve segmentation metrics up to 0.69 on whole-slide images.
• The study highlights challenges in distinguishing benign from normal tissues and underscores the need for robust, interpretable digital pathology tools.

Overview of the BACH Challenge on Breast Cancer Histology Images

The paper "BACH: Grand Challenge on Breast Cancer Histology Images" provides a comprehensive analysis of a large-scale competition organized to advance automated classification in breast cancer histology images. The challenge was conducted in conjunction with the 15th International Conference on Image Analysis and Recognition (ICIAR 2018). The primary objective was to evaluate and improve automatic classification algorithms using hematoxylin-eosin stained histopathological images.

Objectives and Methodology

The challenge focused on two tasks:

1. Classification of Microscopy Images: Participants were required to classify images into four categories: Normal, Benign, In situ carcinoma, and Invasive carcinoma.
2. Segmentation of Whole-Slide Images (WSI): This task involved pixel-wise labeling of WSI into the same four categories.

The dataset provided for the competition included both images from microscopy and whole-slide scans, aiming to address variability across different acquisition methods.

Approaches and Results

The challenge attracted significant participation with 64 submissions from 677 registrations. Various convolutional neural networks (CNNs) were employed by participants, with notable architectures including VGG, Inception, ResNet, and DenseNet. These models were often pre-trained on ImageNet to compensate for limited training data, leveraging transfer learning to fine-tune the models for medical image classification.

• Classification Task: The best-performing submissions achieved an accuracy of around 87% for classifying the microscopy images. This performance was comparable to human pathologists, indicating the capability of CNNs to handle breast cancer classification tasks effectively.
• Segmentation Task: For WSI analysis, the submissions achieved a maximum score of 0.69 for the custom evaluation metric designed to assess the segmentation performance.

Key Findings and Challenges

1. Integration of Contextual Information: Successful approaches frequently integrated both local (microscopic) and global (tissue architecture) context. Large receptive fields were favored to capture this information.
2. Performance Variability: Despite high average accuracy, the methods showed difficulty with certain classes, particularly distinguishing benign from normal tissues. This highlights the need for improved feature extraction and representation techniques.
3. Comparative Evaluation: The study included an inter-observer analysis, where the performance of top AI solutions was compared against experienced pathologists, demonstrating competitive performance.
4. Challenges in Generalization: Differences between cross-validation accuracy and independent test set results underscored the challenges of generalization, heavily influenced by variability in staining and acquisition.

Implications and Future Directions

The BACH challenge highlighted both the potential of deep learning solutions in histopathological image analysis and the gaps needing further research, such as developing methodologies that provide interpretability alongside high classification accuracy. Future work should focus on enhancing model robustness, integrating more diverse datasets for training, and addressing ethical considerations for clinical applications.

Overall, BACH has contributed significantly to the field by providing a public, annotated dataset and a benchmark for future developments in digital pathology. The outcomes suggest a promising path towards automated diagnostic tools that can assist pathologists in clinical settings, ultimately aiding in better management of breast cancer.