Women With Breast Cancer
Conditions
Brief summary
This project aims to establish a comprehensive artificial intelligence system for detecting and qualitative diagnosing breast lesions. Mammary images will be used to construct a diagnosis method based on deep learning. The system is proposed to automatically analyze the type of mammary glands, automatically identify and mark all breast lesions on the mammography images, provide the malignancy probability judgment of the lesions, the BI-RADS classification and the clinical suggestion, and also automatically generate the structured diagnosis report.
Detailed description
This is a multi-center study.The project contains a retrospective part(3000 samples anticipated) and a prospective part(7000 samples anticipated). In the retrospective part, investigators collected subjects with mammary images to design the deep learning method and construct a detective and diagnostic model for breast lesions. In the prospective part, investigators validate the accuracy of the constructed deep learning method, and established artificial intelligence system focusing on mammary diagnosis. Investigators will also explore the application pattern of the artificial intelligence system in clinical practice.
Interventions
DIAGNOSTIC_TESTmammography
When a woman comes to the clinic to receive mammography. Then a radiologist will give a BI-RADS classification after reviewing the images. If a BI-RADS 4/5 is obtained, the woman will receive pathological biopsy to ensure there is a benign or malignant lesion. If a BI-RADS 3 is obtained, the woman will be followed up by a half-year interval until two year after the first mammography. At each follow up, she will receive mammography. If a BI-RADS 4/5 is obtained at follow up, she will receive pathological biopsy; if a BI-RADS 1/2/3 is obtained at follow up, she will be followed up by a half-year interval until two year. If a BI-RADS 1/2 is obtained at the first mammography, the woman will receive a second mammography after two year. During the study period, breast examination and results will be recorded for every subject. Radiologists will give the diagnosis with and without AI support.
Sponsors
Peking University
Peking University Cancer Hospital & Institute
Study design
Observational model
COHORT
Time perspective
OTHER
Eligibility
Sex/Gender
FEMALE
Age
18 Years to No maximum
Healthy volunteers
No
Inclusion criteria
* the X-ray images of the breast were complete * the results of pathological diagnosis or more than 2 years of mammography follow-up were available * subject signs informed consent(this item was only for prospective study cases)
Exclusion criteria
* there exists pathological diagnosis of breast lesions when receiving mammography * there lacks pathological diagnosis or 2 years of mammography follow-up * subject withdraws(this item was only for prospective study cases)
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| benign-malignant diagnosis accuracy | from the first mammography to pathological result obtained(an average of 3 weeks if mammography BI-RADS 4 or 5 obtained) | the accuracy of the AI model, radiogist with AI support, radiologist alone for binary diagnosis of a benign or malignant breast lesion according to pathology. If either one mammography of BI-RADS 4/5 in the first examination or during the two year' follow up examination is obtained,a pathological examination is performed, the lesion is judged benign or malignant according to pathological results. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| lesion detection accuracy | from the first mammography to radiologist diagnosis (within 3 days after the mammography taken) | the detection rate of the constructed deep learning method for detecting benign or malignant breast lesion according to radiologist's subjective diagnosis or follow up as reference. If a radiologist suggests existence of a lesion at the first mammography or at each follow-up mammography during the 2-year period, it is considered that a lesion exists |
Countries
China
Outcome results
None listed