Augmentation-based deep learning for identification of circulating tumor cells.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Oncology

Background:

• Circulating Tumor Cells (CTCs) are vital liquid biopsy biomarkers for noninvasive cancer management.
• Identifying CTCs is challenging due to low numbers, heterogeneity, and limitations of fluorescence-based methods.
• Manual analysis of single-cell images is time-consuming and variable.

Purpose of the Study:

• To develop a Deep Learning (DL) pipeline for distinguishing CTCs from leukocytes using bright-field images.
• To enhance diagnostic accuracy and optimize clinical workflows for CTC analysis.
• To overcome limitations of fluorescence-based methods by utilizing bright-field imaging.

Main Methods:

• Utilized Parsortix® and DEPArray™ technologies for unbiased CTC isolation and single-cell visualization.
• Developed a ResNet-based Convolutional Neural Network (CNN) for image classification.
• Applied data augmentation and incorporated fluorescence (DAPI) channel images during training for improved generalization.

Main Results:

• The DL model achieved an F1-score of 0.798 in distinguishing CTCs from leukocytes.
• The model successfully identified CTCs using only bright-field images during testing.
• Demonstrated the potential of DL to refine CTC analysis without fluorescence markers.

Conclusions:

• Deep learning offers a powerful tool for accurate CTC identification in liquid biopsies.
• Bright-field single-cell analysis combined with DL can overcome current limitations in CTC detection.
• This approach has the potential to significantly advance cancer patient management and diagnostics.