Detection of Breast Cancer Lump and BRCA1/2 Genetic Mutation under Deep Learning.

Abstract

To diagnose and cure breast cancer early, thus reducing the mortality of patients with breast cancer, a method was provided to judge threshold of image segmentation by wavelet transform (WT). It was used to obtain information about the general area of breast lumps by making a rough segmentation of the suspected area of the lump on mammogram. The boundary signal of the lump was obtained by region growth calculation or contour model of local activity. Meanwhile, multiplex polymerase chain reaction (mPCR) and mPCR-next-generation sequencing (mPCR-NGS) were used to detect BRCA1/2 genome. Sanger test was used for newly high virulent mutations to verify the correctness of mutagenic sites. The results were compared with the information marked by experts in the database. According to Daubechies wavelet coefficients, the average measurement accuracy was 92.9% and the average false positive rate of each image was 86%. According to mPCR-NGS, there was no pathogenic mutation in the 7 patients with high-risk BRCA1/2 genetic mutations. Single nucleotide polymorphism (SNP) in nonsynonymous coding region was detected, which was consistent with the Sanger test results. This method effectively isolated the lump area of human mammogram, and mPCR-NGS had high specificity and sensitivity in detecting BRCA1/2 genetic mutation sites. Compared with traditional Sanger test and target sequence capture test, it also had such advantages as easy operation, short duration, and low cost of consumables, which was worthy of further promotion and adoption.