Ultrasound medical image analysis has for a long time been limited by the size of the dataset and has often struggled to replenish the modal characteristics of ultrasound image data. Also, most of the work has focused on the downstream tasks of automatic diagnosis while ignoring the complementarity of incremental features from standard cut views of normal populations. Therefore, this study addresses these issues by providing a large-scale multitasking ultrasound image dataset, which is the largest known multitasking ultrasound dataset with a total of 120,354 data. The UMTD was collected from real patients in Quanzhou, China, with a total of 3600 independent patients. The UTMD dataset consists of two parts that have been randomly partitioned. One part is the train file for training and the other part is the val file for data testing. There are 105,309 data for training and 15,045 data for testing. There will be 28 category folders in the train or val folder, and these category folders make up the data category annotation of UMTD. In each category folder, all 2D ultrasound image data are provided in 512*512 pixels, and each image is named to show the category of the ultrasound image and the selected offline augmentation method. In addition, we provide a new self-supervised learning method CDOA for the modality of ultrasound data, which can achieve satisfactory results without any online augmentation while adaptively learning ultrasound image characteristics. We hope that the UMTD data will bring a fundamental contribution of data support to the ultrasound image analysis community, especially to fill the task gap of lacking large-scale datasets for parameter pre-training on ultrasound data modalities. We also hope that the CDOA self-supervised approach to understanding ultrasound images from a frequency perspective will bring a new perspective on ultrasound data analysis.Detailed information about the code can be found on:https://github.com/JsongZhang/CDOA-for-UMTD

Public access to UMTD data only occurs after NeurIPS 2023 reception.