Artificial Intelligence for CELLSEARCH Image Analysis.

Area of Science:

• Oncology
• Biomarker Discovery
• Artificial Intelligence in Medicine

Background:

• Circulating tumor cells (CTCs) and tumor-derived extracellular vesicles (tdEVs) are key prognostic indicators in metastatic carcinomas.
• Manual analysis of CTCs and tdEVs from CellSearch images is subjective and labor-intensive.
• Standardization across research centers is crucial for reliable prognostic value.

Purpose of the Study:

• To develop and evaluate AI-based methods for automated analysis of CellSearch images for CTC and tdEV identification.
• To introduce Contrast Maximization (CM) software for automated CTC and tdEV quantification.
• To establish Blood Tumor Load (BTL) as a combined biomarker integrating CTC and tdEV counts.

Main Methods:

• Development of AI-based software (Contrast Maximization - CM) for automated analysis of CellSearch immunofluorescent images.
• Evaluation of CM for identifying CTCs and tdEVs in metastatic breast, colorectal, and prostate cancer.
• Creation and validation of the Blood Tumor Load (BTL) biomarker by combining CM-derived CTC and tdEV counts.

Main Results:

• CM-based CTC identification (CM-CTC) showed performance comparable to or exceeding human operators in predicting patient survival.
• CM-based tdEV classification (CM-tdEV) outperformed traditional human-designed gating strategies.
• The combined BTL biomarker, particularly when derived from CM, demonstrated superior prognostic value compared to individual CTC or tdEV counts.

Conclusions:

• AI-based automated analysis using CM enhances the accuracy, speed, and reproducibility of CTC and tdEV quantification.
• The Blood Tumor Load (BTL) biomarker, leveraging AI, provides a more robust prognostic predictor than CTCs or tdEVs alone.
• Automated and standardized analysis of CTCs and tdEVs via CM strengthens their clinical utility in metastatic cancer management.