Artificial Intelligence for CELLSEARCH Image Analysis

The number of circulating tumor cells (CTC) and tumor-derived extracellular vesicles (tdEV) is an independent predictor of survival in patients with metastatic carcinomas. Being one order of magnitude more abundant than CTC, tdEV can provide complementary prognostic value. On CellSearch immunofluorescent images, operators can identify CTC and tdEV based on morphology, DNA, cytokeratin, and CD45 fluorescent intensity. This manual process is time-consuming and potentially affected by subjective interpretations. To overcome these limitations and maximize standardization among different research centers, we introduce artificial intelligence (AI) based methods for the automated analysis of CellSearch images. These methods were evaluated on fluorescent images of metastatic breast, colorectal, and prostate cancer studies. This chapter will present Contrast Maximization (CM), an AI-based software solution for automated CTC and tdEV identification, and the concept of Blood Tumor Load (BTL), which combines CTC and tdEV counts into a single interpretable biomarker ranging from 0 (favorable) to 1 (unfavorable). CM-CTC demonstrated performance comparable to or better than human operators in predicting overall survival in metastatic cancer patients, enabling rapid and reproducible enumeration of clinically relevant CTCs. Similarly, CM-tdEV outperformed tdEV classification based on human-designed gating strategies. Survival analysis further revealed that BTL outperformed CTC and tdEV counts alone, underscoring the added value of combining these two complementary biomarkers. Notably, BTL derived from CM outperformed BTL based on manual counts, highlighting the advantages of CM in automating and standardizing the analysis process, ultimately enhancing accuracy and strengthening the relationship with clinical outcomes.