Circulating tumor cells (CTCs) enumeration and machine-learning based diagnostic biomarkers for breast cancer detection.

[BACKGROUND] Circulating tumor cells (CTCs) are detectable in early-stage cancer and may enable early cancer detection. We evaluated a CTC-based assay as a complementary biomarker for breast cancer detection in an Asian population with a high prevalence of dense breast tissue.

[METHODS] In this single-center, prospective, blinded study, peripheral blood from Taiwanese women with breast cancer and healthy controls was analyzed using a CTC-enumeration platform (CMx) based on biomarker expression (cytokeratin 18 [CK18], mammaglobin [MGB], CD45), cell morphometry, and nuclear features. A machine-learning model integrating CTC biomarkers with age, white blood cell (WBC) count, and platelet count was developed to assess classification performance, providing proof-of-concept for combining CTC-derived and routine blood parameters in breast cancer risk assessment.

[RESULTS] A total of 228 breast cancer patients and 170 healthy controls were included. Age and CK18- and MGB-positive CTC counts differed significantly between groups, whereas WBC and platelet counts did not. An ensemble linear support vector machines model incorporating age and CTC features achieved an area under the curve of 0.85 (95% CI, 0.73-0.96) in the independent test cohort, with high sensitivity (0.93), positive predictive value (0.74), and negative predictive value (0.86), but modest specificity (0.57). In the exploratory BI-RADS 3/4 subgroup, the model identified all cancer cases (sensitivity 1.00), with a specificity of 0.44 and overall accuracy of 0.79.

[CONCLUSIONS] This study demonstrates the feasibility of combining CTC enumeration with machine learning for breast cancer detection and supports the need for future large-scale, multicenter, multiethnic prospective external validation.