A breast cancer detection method integrating deep multi-instance learning and inter-package similarity

Abstract

A breast cancer detection method that fuses deep multi-instance learning and inter-package similarity. Involving the fields of pattern recognition, image processing and computer vision, this method studies the problem that the similarity cannot be automatically learned in the deep multi-instance learning method based on the similarity between packets and the application of the method in breast cancer detection. Existing deep multi-instance learning methods based on inter-packet similarity have good results in breast cancer detection, but this method uses a fixed similarity measure and does not consider its automatic learning problem. The similarity between packages is automatically learned, and the optimized network can automatically learn the more important relationships between packages and examples. Compared with the original method, the breast cancer detection accuracy based on this method is higher and has certain practical value.

Description

technical field [0001] The invention relates to the fields of pattern recognition, image processing, computer vision and the like, in particular to a breast cancer detection method based on deep multi-example learning. Background technique [0002] In recent years, artificial intelligence has played an increasingly important role in medical auxiliary diagnosis. It can not only assist in tedious patient screening work, but also avoid subjective errors caused by manual reading to a certain extent. However, it is often difficult to obtain fine-grained labels (eg, pixel-level labels) for medical images, which makes traditional strongly supervised learning methods difficult to apply in medical image processing. In order to solve the above problems, weakly supervised learning techniques that only use whole image annotations have been widely used in medical aided diagnosis and have achieved comparable or better results than strongly supervised learning. As a typical weakly supervi...

AI Technical Summary

Problems solved by technology

However, it is usually very difficult to obtain fine-grained labels (such as: pixel-level labels) of medical images, which makes traditional strong supervised learning methods difficult to apply in medical image processing.

Images