Circulating tumor cells (CTCs) are promising biomarkers for cancer diagnosis, while detecting CTCs in clinical samples is still challenging due to the scarcity and heterogeneity of CTCs. Herein, a triple-mode sensing platform based on an antifouling capture and dual-targeting recognition strategy is presented for sensitive detection of human cervical cancer cells (HeLa). First, the target CTCs can be specifically captured using antifouling magnetic nanoparticles (AMNPs) encoded with AS1411 aptamers (AMNPs-AptAS1411), which not only enhance the capture efficiency for CTCs but also reduce the nonspecific adsorption of free proteins. To further improve the detection specificity, a dual-targeting recognition system is constructed in which the aptamer MUC1 (AptMUC1) and cholesteryl-modified helper DNA (chol-Helper) can recognize MUC1 proteins and membrane phospholipids, respectively. The proximity hybridization of AptMUC1 and chol-Helper triggers the structure switching of chol-Helper to release the initiator for the hybridization chain reaction (HCR), in which G-quadruplex polymer chains are generated for signal amplification. By virtue of biocatalysis of G-quadruplex/hemin DNAzyme, a triple-mode sensing platform is constructed using absorbance/colorimetric/photothermal signal readout, exhibiting an ultrahigh sensitivity as low as 5 cells/mL. Moreover, taking advantage of the triple-mode readout as sensing units, a sensor array is established combined with a machine learning algorithm to distinguish five kinds of CTCs, which holds significant potential for early diagnosis and classification of cancers.