Improved terraced compression method combined with BIRCH algorithm for heterogeneity detection

Breast cancer screening based on multispectral transmission imaging has the advantages of being harmless and easy to implement. If combined with image processing means, automated breast heterogeneity detection can be realized, which will create great value. Due to the limitation of the amount of data as well as labeled data, it is not possible to use supervised learning methods for classification and detection, while clustering methods are more appropriate. However, due to the strong absorption as well as scattering properties of biological tissues, the signal-to-noise ratio of the acquired breast transmission image is extremely low. To improve the image, frame accumulation techniques are often used, but this will cause data redundancy, making existing clustering methods ineffective in distinguishing different substances in such images. In order to realize the detection of heterogeneity in transmission images of breasts, an improved terraced compression method combined with the BIRCH algorithm is used to process the images. Through the image acquisition and processing of the phantom, the dice index and purity of the image are 0.8 and 0.81 respectively. The results show that the different substances in the phantom can be well detected and identified by the method, which provides a reference for the early screening of breast cancer.