The purpose of this study is using a deep learning method to analyze the automated breast ultrasound (ABUS) and hand-held ultrasound(HHUS) images, establish and evaluate a diagnosis, therapy assessment and prognosis prediction model of breast cancer. The model would provide important references for further early prevention, early diagnosis and personalized treatment.

1. Establishing a database By collecting ABUS, HHUS and comprehensive breast images data, essential information, clinical treatment information, prognosis, and curative effect information, a complete breast image database is constructed.
2. Marking ABUS images Three doctors use a semi-automatic method to frame the lesions on the image.
3. Building the model Using the deep learning method to preprocess, analyze and train the marked images, and finally get a model diagnosis, efficacy evaluation and prognosis prediction model of breast cancer.
4. Evaluating the model 1）Self-validation： Analyze the sensitivity, AUC of the breast cancer diagnosis model and the false-positive number on each ABUS volume.

2) Compared the sensitivity, AUC and the false-positive number with a commercial diagnosis model.

3)To test the screening and diagnostic efficacy of computer-aided diagnosis systems through prospective or retrospective studies.

4)By analyzing the size and characteristics of the lesions after neoadjuvant chemotherapy, and predicting the OS and DFS time, the therapy assessment and prognosis prediction model were evaluated.

Inclusion Criteria:

1. Female patients over 18 years old who come to the two centers for physical examination or treatment;
2. Complete basic information and image data

Exclusion Criteria:

1. There is no complete ABUS and HHUS images data;
2. The image quality is poor;
3. In multifocal breast cancer, the correlation between the tumor in the image and the postoperative pathological examination is uncertain.

Female patients over 18 years old from two countries (China and Korea).