Multi-view mammogram analysis method, multi-view mammogram analysis system, and non-transitory computer-readable medium

A multi-view mammogram analysis system, comprising:

• a storage device, configured for storing a plurality of training images and a plurality of mammogram images, wherein the plurality of mammogram images at least include two mammogram images captured by a craniocaudal view and a mediolateral oblique view from the same side;
• and
• a processor, electrically connected to the storage device, the processor comprises:
• an abnormal analyzing circuit which is configured for utilizing a symptom identification model to determine whether the plurality of mammogram images have an abnormal state, and generating a plurality of heat maps corresponding to the plurality of mammogram images;
• a false positive analyzing circuit which is electrically connected to the abnormal analyzing circuit, and configured for utilizing a false positive filtering model to determine whether the plurality of heat maps have a false positive feature, and generating an abnormal probability corresponding to the plurality of heat maps, wherein the false positive filtering model is trained by a plurality of training images captured by the craniocaudal view and the mediolateral oblique view;
• and
• an abnormal position analyzing circuit which is electrically connected to the false positive analyzing circuit, and configured for utilizing a first threshold to determine the abnormal probability, if the abnormal probability is greater than the first threshold, detecting and outputting a lesion position corresponding to the plurality of heat maps.

A non-transitory computer-readable medium including one or more sequences of instructions to be executed by a processor for performing a multi-view mammogram analysis method, wherein the method comprises:

• inputting a plurality of mammogram images, wherein the plurality of mammogram images at least include two mammogram images captured by a craniocaudal view and a mediolateral oblique view from the same side;
• utilizing a symptom identification model to determine whether the plurality of mammogram images have an abnormal state, and generating a plurality of heat maps corresponding to the plurality of mammogram images;
• utilizing a false positive filtering model to determine whether the plurality of heat maps have a false positive feature, and generating an abnormal probability corresponding to the plurality of heat maps, wherein the false positive filtering model is trained by a plurality of training images captured by the craniocaudal view and the mediolateral oblique view;
• and
• utilizing a first threshold to determine the abnormal probability, if the abnormal probability is greater than the first threshold, detecting and outputting a lesion position corresponding to the plurality of heat maps.