Variations in the intracellular expression level of cancer-related microRNAs (miRNAs) are connected with worsening tumor progression. A simple, accurate, and sensitive analytical method for the imaging and detection of intracellular miRNA is still a great challenge due to the low abundance of miRNAs and the complexity of intracellular environments. In this work, target miRNA (miRNA)-mediated catalytic hairpin assembly (CHA)-induced gold nanocage (GNC)-hairpin DNA1 (hpDNA1)-hpDNA2-GNC nanostructures were designed for surface-enhanced Raman scattering (SERS) detection and imaging of the specific miR-125a-5p in the normal lung epithelial cell line (BEAS-2B cells) and lung cancer cell line (A549 cells). The finite difference time domain (FDTD) simulations showed that the polymer of GNCs possessed a much stronger electromagnetic field in nanogaps than that of single GNC, theoretically confirming the rational design of the CHA assembly strategy. Using this method, miR-125a-5p can be detected in a wide linear range with a detection limit of 43.96 aM and high selectivity over other miRNAs in vitro. Moreover, SERS imaging successfully detected and distinguished the expression levels of intracellular miR-125a-5p in BEAS-2B cells and A549 cells. The results obtained by the SERS assay were consistent with those obtained by the real-time quantitative polymerase chain reaction (qRT-PCR). This method can offer a powerful strategy for the imaging and quantitative detection of various types of biomolecules in vitro as well as in living cells.