A Pulmonary Ultrasound Dataset for Tumor Segmentation and Classification

An open-access ultrasound image dataset for Peripheral pulmonary lesions (PPLs) segmentation and classification is proposed, which contains 2,756 B-mode ultrasound images of pulmonary lesions, as well as complete annotations and biopsy confirmations of the pulmonary lesions from expert sonographers. More detailed, the dataset was partitioned into five folds using a stratified cross-validation strategy, ensuring that the proportions of benign and malignant cases were approximately balanced in both the training and testing sets of each fold.