The PANDA challenge

前列腺癌分级评估 (PANDA) 挑战

> 熊猫挑战赛已经结束。我们目前正在更新网站。

每年报告的新诊断超过100万例，前列腺癌 (PCa) 是全球男性中第二大最常见的癌症，每年导致350,000例以上的死亡。降低死亡率的关键是开发更精确的诊断方法。PCa的诊断基于前列腺组织活检的分级。这些组织样本由病理学家检查，并根据格里森分级系统进行评分。在这个挑战中，您将开发用于检测前列腺组织样本图像上的PCa的模型，并使用格里森分级上最广泛的多中心数据集来估计疾病的严重程度。

分级过程包括根据肿瘤的结构生长模式发现癌组织并将其分类为所谓的格里森模式 (3、4或5) (图1)。在给活检分配格里森评分后，将其转换为1-5等级的ISUP等级。格里森分级系统是PCa最重要的预后指标，ISUP分级在决定如何治疗患者时起着至关重要的作用。既有癌症缺失的风险，也有导致不必要治疗的过度风险。但是，该系统在病理学家之间存在明显的观察者间差异，从而限制了其对个别患者的有用性。评分的这种差异可能导致不必要的治疗，或者更糟的是，错过了严重的诊断。

自动化深度学习系统在准确分级PCa方面显示了一些希望。最近的研究，包括由主持这一挑战的小组独立进行的两项研究，表明这些系统可以实现病理学家水平的表现。然而，这些系统/结果没有在规模上使用多中心数据集进行测试。

您在这里的工作将使用格里森分级上最广泛的多中心数据集来改进这些工作。训练集包括大约11,000张来自两个中心的数字化H & E染色活检的全幻灯片图像。这是可用的最大的公共全幻灯片图像数据集，大约是CAMELYON17挑战的8倍，这是该领域最大的数字病理学数据集之一，也是最著名的挑战。此外，与以前的挑战相反，我们正在提供完整的诊断活检图像。使用相当大的多中心测试集，由专业的uro病理学家分级，我们将评估挑战提交的适用性，以改善这一关键诊断功能。

Prostate Cancer Grading Assessment (PANDA) Challenge

The PANDA Challenge has concluded. We are currently updating our website.

Prostate cancer (PCa) is the second most common cancer among men globally, with over 1 million new diagnosed cases reported annually and more than 350,000 deaths each year. The key to reducing mortality lies in developing more accurate diagnostic methods. The diagnosis of PCa is based on the grading of prostate tissue biopsies. These tissue samples are examined by pathologists and scored according to the Gleason grading system. In this challenge, you will develop models for detecting PCa on images of prostate tissue samples, and estimate disease severity using the largest multi-center dataset to date for Gleason grading.

The grading process involves identifying cancerous tissue based on the structural growth pattern of tumors and classifying it into the so-called Gleason patterns (3, 4, or 5) (Figure 1). After assigning a Gleason score to a biopsy, it is converted to an ISUP grade ranging from 1 to 5. The Gleason grading system is the most important prognostic indicator for PCa, and ISUP grades play a critical role in deciding how to treat patients. There are risks of both missing cancer and causing unnecessary treatment via overgrading. However, this system exhibits significant inter-observer variability among pathologists, which limits its utility for individual patients. Such variability in scoring can lead to unnecessary treatment or, worse, missed critical diagnoses.

Automated deep learning systems have shown some promise in accurately grading PCa. Recent studies, including two independent studies led by the group organizing this challenge, have demonstrated that these systems can achieve pathologist-level performance. However, these systems/results have not been tested at scale using multi-center datasets.

Your work here will build upon these efforts using the largest multi-center dataset for Gleason grading to date. The training set includes approximately 11,000 whole-slide images of digitized hematoxylin and eosin (H&E)-stained biopsies from two centers. This is the largest public whole-slide image dataset available, approximately 8 times the size of the CAMELYON17 Challenge, one of the largest digital pathology datasets and most prominent challenges in the field. Furthermore, unlike previous challenges, we are providing full diagnostic biopsy images. Using a large, multi-center test set graded by specialized uro-pathologists, we will evaluate the suitability of challenge submissions to improve this critical diagnostic function.