Small extracellular vesicles (sEVs) derived from tumors contain a vast amount of cellular information and are regarded as a potential diagnostic biomarker for noninvasive cancer diagnosis. Nevertheless, it remains challenging to accurately measure sEVs from clinical samples due to the low abundance of these vesicles as well as their phenotypic heterogeneity. Herein, a polymerase-driven logic signal amplification system (PLSAS) was developed for the high-sensitivity detection of sEV surface proteins and breast cancer (BC) identification. Aptamers were introduced to serve as sensing modules to specifically recognize target proteins. By changing the input DNA sequences, two polymerase-driven primer exchange reaction systems were rationally designed for DNA logic computing. This allows for autonomous targeting of a limited number of targets using "OR" and "AND" logic, leading to a significant increase in fluorescence signals and enabling the specific and ultrasensitive detection of sEV surface proteins. In this work, we investigated surface proteins of mucin 1 (MUC1) and the epithelial cell adhesion molecule (EpCAM) as model proteins. When MUC1 or EpCAM proteins were used as single signal input in the "OR" DNA logic system, the detection limit of sEVs was 24 or 58 particles/μL, respectively. And MUC1 and EpCAM proteins of sEVs can be simultaneously detected in the AND logic method, which can significantly reduce the effect of phenotypic heterogeneity of sEVs to distinguish the source of sEVs derived from various mammary cell lines, such as MCF-7, MDA MB 231, SKBR3, and MCF-10A. The approach has achieved high discrimination in serologically tested positive BC samples (AUC 98.1%) and holds significant potential in advancing the early diagnosis and prognostic assessments of BC.