This study aims to use a combination of abbreviated protocol and artificial intelligence to automatically identify lesions and make diagnosis without decreasing the diagnostic accuracy of breast cancer, thus enhancing the comfort of patient examination, accelerating the flow of examination and reducing the load of clinical work.

Full sequence of MRI scan is:

MR scan protocol:

1. Magnetic fields and coils: Use MR scanners with high fields of 1.5 T and above with dedicated breast coils.
2. Scanning position: prone position with bilateral breasts naturally draped over the center of the breast coil. The position should ensure that all breast tissue is located in the coil, the skin and breast are not folded, the bilateral breast is symmetrical, the nipple is perpendicular to the ground, and the midline of the sternum is located in the middle line of the coil.
3. Scanning sequence and parameters: T1WI non-fat suppressed sequence, T2WI fat suppressed sequence, dynamic five-phase enhanced T1WI fat suppressed sequence, and diffusion weighted imaging (DWI); delayed sagittal T1W1 fat suppressed sequence.

Imaging parameters: the thickness of the scanned layer should be ≤3 mm, the resolution within the layer should be <1.5 mm, and the single scan time should be <2 min.

Enhancement scan: Gadobutrol used as the contrast agent, and the injection dose was 0.1 ml/kg, which was injected through the elbow vein at a rate of 2-3 mL/s using a pressure syringe, and 10-20 mL of saline was injected into the tube at the same rate after the contrast agent injection. The T1WI sequences before and after enhancement were preferably fat-suppressed and bilateral mammary glands were imaged simultaneously, and subtraction treatment was recommended. The recommended duration of delayed enhancement scan is 7 min, not less than 5 min.

Abbreviated sequences : T1WI + dynamic enhanced T1 phase I + maximum density projection images generated by automatic reconstruction in three directions. No extra sequences are required.

By adding artificial inteeligence, a diagnostic performance comparable with full sequences is expected.

Inclusion Criteria:

1. Patients with breast lesions detected by ultrasound and mammography that cannot be characterized
2. Patients who were consecutively included in our hospital for breast MRI without treatment
3. Underwent preoperative full-protocol breast MRI
4. Pathological results are available, of which benign lesions can be determined by follow-up

Exclusion Criteria:

1. Poor MRI image quality
2. Patients who have been received the biopsy

The statistical analyses in this study were all descriptive, without any predefined hypotheses.

Referring to a recent study* of artificial intelligence combined with abbreviated protocol MRI to classify benign and malignant non-mass-enhancing lesions in the breast*, a sample size of 800 cases was taken, which would require 800 patients to be enrolled, based on at least one lesion at the initial consultation of one patient.