Since July 2021 the TUPAC challenge is hosted on this platform. The original website hosting the challenge is no longer available. 

If you use the dataset from this challenge please cite the challenge overview paper:

Veta, Mitko, et al. "Predicting breast tumor proliferation from whole-slide images: the TUPAC16 challenge." Medical image analysis 54 (2019): 111-121.

Introduction

The tumor proliferation speed (tumor growth) is an important biomarker indicative of breast cancer patients' prognosis. Breast cancer patients with high tumor proliferation speed have worse outcomes compared with patients with low tumor proliferation speed. Thus, the assessment of this biomarker influences the decisions for the treatment plan of the patient (patients with aggressive tumors can be treated with aggressive therapy).

Tumor proliferation in a clinical setting is assessed by pathologists. The most common method is by counting of mitotic figures (dividing cell nuclei) in hematoxylin & eosin (H&E) stained histological slide preparations that are examined under a microscope. Large density of mitotic figures indicates high tumor proliferation speed. Although this task is routinely performed in almost every pathology practice, mitosis counting is known to suffer from reproducibility problems that are caused by the underlying subjectivity of the process. The state of the art in automatic mitosis detection, which is seen as a solution to the problem of subjectivity, has been significantly advanced in recent years. This has been in large part the result of the organization of challenges on this topic.

Although state-of-the-art mitosis detection methods approach the performance of human observers, this is achieved in more controlled conditions when mitosis detection is performed in pre-selected regions of relevant tumor tissue. In a more practical scenario, however, mitosis detection must be applied to whole-slide images (WSIs). In addition, the results from the detection of mitotic figures must be summarized in a proliferation score that can be integrated in current prognostic grading systems. 

With this challenge we aim to address these limitations and further advance the state of the art in automatic tumor proliferation scoring from whole slide images. 

Challenge goal and dataset

The participants of this challenge will be provided with a dataset of whole slide images with known tumor proliferation scores. The goal of the challenge is to assess algorithms that predict the tumor proliferation scores from the whole slide images.

In addition, two auxiliary datasets will be provided: 1) a dataset with annotated mitotic figures that can be used to train a mitosis detection method, and 2) a dataset with annotations of regions of interest that can be used to train a region of interest detection method.