Circulating Tumor Cell Identification Based on Deep Learning.

Area of Science:

• Oncology
• Biomedical Engineering
• Computational Biology

Background:

• Circulating tumor cells (CTCs) are crucial biomarkers for cancer metastasis, diagnosis, and prognosis.
• CTC detection via blood tests offers safety, low cost, and repeatability but is labor-intensive and prone to errors.
• Automated CTC detection is needed to improve accuracy and efficiency in clinical settings.

Purpose of the Study:

• To develop and validate a novel convolutional neural network (CNN) method for automated detection of CTCs.
• To assess the performance of CNN and transfer learning models in identifying CTCs from immunofluorescence in situ hybridization (imFISH) images.
• To evaluate the clinical utility of AI-driven CTC detection for cancer prognosis and metastasis assessment.

Main Methods:

• Collected peripheral blood from 776 cancer patients.
• Enriched CTCs using Cyttel, removing leukocytes.
• Identified CTCs using imFISH with CD45+, DAPI+, and CEP8+ staining.
• Developed and applied traditional CNN and transfer learning models for CTC detection.

Main Results:

• The traditional CNN model achieved 95.3% sensitivity and 91.7% specificity.
• The transfer learning model demonstrated higher performance with 97.2% sensitivity and 94.0% specificity.
• Both CNN models effectively detected CTCs with high sensitivity.

Conclusions:

• The developed CNN and transfer learning methods offer accurate and automated CTC detection.
• These AI-based approaches have significant clinical reference value for prognosis and metastasis diagnosis.
• Automated CTC detection can enhance the efficiency and reliability of cancer biomarker analysis.