The invention relates to a
breast cancer image
anomaly detection method and
system based on self-
supervised learning, and the method comprises the steps: based on the
biophysics principle, including
cell fluctuation dynamics and a tumor
stress field model, generating various types of synthetic anomaly images through the
simulation of a
normal breast ultrasound image; the normal image and the composite image are jointly input into a
feature extraction network fusing Transform and KAN for self-supervised contrast learning pre-training; after training is completed, the model is used for extracting multi-scale features of normal images to construct a normal feature
memory bank; a local abnormal
score is obtained by calculating the
similarity distance between the to-be-detected image features and the features in the
memory bank, so that accurate abnormal region recognition and positioning without
manual annotation are realized; and finally, carrying out visual output on a detection result. According to the method, the problems of high dependency of
labeled data, insufficient sensitivity of
small lesion detection and the like are effectively solved, and the
automation level of
breast cancer screening and the reliability of clinical auxiliary diagnosis are remarkably improved.