Machine Learning Approach for Enumeration of Circulating Cells with Diffuse in vivo Flow Cytometry.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Cancer Research

Background:

• Diffuse in vivo flow cytometry (DiFC) is an emerging technique for enumerating rare circulating tumor cells (CTCs) in small animals non-invasively.
• Existing amplitude threshold-based methods for DiFC analysis have limitations in distinguishing CTC signals from instrument noise and artifacts, potentially reducing detection efficiency.

Purpose of the Study:

• To develop and validate a machine learning (ML)-integrated signal processing approach for enhanced CTC enumeration using DiFC.
• To improve the accuracy and robustness of CTC detection by distinguishing true CTC signals from artifacts based on peak characteristics.

Main Methods:

• An ML-integrated approach utilizing a convolutional neural network (CNN) classifier was developed.
• The CNN was trained to analyze both peak amplitude and temporal shape characteristics for CTC identification.
• The model's performance was rigorously validated using in-silico, control, and CTC-bearing mouse datasets.

Main Results:

• The CNN classifier demonstrated high performance, achieving accuracy, precision, sensitivity, and specificity exceeding 98% on test data.
• The ML-integrated method significantly increased the accurate identification of CTCs and their flow direction compared to the previous threshold-based approach.
• False positive detections were substantially reduced across all validation datasets, indicating improved signal processing.

Conclusions:

• The ML-integrated approach represents a significant advancement in DiFC-based CTC enumeration.
• This method enhances robustness against artifacts, particularly in noisy experimental conditions.
• The improved CTC detection capabilities hold promise for more effective non-invasive cancer monitoring and research in small animal models.