FedSurg2024

该数据集包含来自 4 个不同中心的阑尾切除术图像。每个中心都用关键帧和相应的阑尾炎腹腔镜分级注释了他们的视频。以 2 FPS 的帧速率在关键帧周围 +/-50 秒内采样，从每个视频中提取了 200 帧。第 100 帧是带注释的关键帧。一些视频是立体捕捉的。来自三个训练中心的数据集被分为三个主要部分：训练公共（16%），训练私人（64%），测试（20%）。来自第四个中心的数据集用于测试。可以访问数据集的训练公共部分。这意味着有 16% 的数据来开发方法。提交后的方法将在完整的训练集（训练公共 + 训练私人 [80% 的数据]）上进行训练，并在测试集（20% 的数据）和第四个中心的数据上进行测试。可以决定是否只使用帧 100（关键帧）、数据的一部分或完整的 200 帧进行训练。This dataset comprises appendectomy images sourced from four distinct medical centers. Each center annotated its laparoscopic videos with key frames and corresponding laparoscopic grading labels for appendicitis. For each video, 200 frames were extracted via sampling within a ±50-second window surrounding the key frame at a frame rate of 2 FPS, with the 100th frame being the annotated key frame. Some videos were captured stereoscopically. The dataset from the three training centers is partitioned into three primary splits: public training (16%), private training (64%), and test (20%). The dataset from the fourth center is reserved exclusively for final testing. Access to the public training split is granted, meaning 16% of the total dataset is available for method development. Submitted models will be trained on the full training set (public training + private training, which constitutes 80% of the total dataset), and evaluated on both the test split (20% of the total dataset) and the dataset from the fourth center. Participants may choose to use only the 100th frame (the key frame), a subset of the data, or the full set of 200 frames for model training.