Single Vesicle Surface Protein Profiling and Machine Learning-Based Dual Image Analysis for Breast Cancer Detection

Area of Science:

• Biomedical Engineering
• Oncology
• Nanotechnology

Background:

• Single-vesicle molecular profiling of extracellular vesicles (EVs) is crucial for cancer detection and monitoring.
• Dual imaging single-vesicle technology (DISVT) offers a method to quantify targeted EVs, but analysis challenges persist due to false signals and large data volumes.

Purpose of the Study:

• To develop a fully automatic, machine learning-based dual imaging analysis method for EVs.
• To apply this AI-assisted DISVT for detecting and staging HER2-positive breast cancer.

Main Methods:

• A convolutional neural network (Resnet34) with transfer learning was employed for image analysis.
• The model was trained using a combination of experimental and synthetic data.
• Fractions of EpCAM- and CD24-positive EVs were quantified in plasma samples from breast cancer patients and healthy donors.

Main Results:

• EpCAM-positive and CD24-positive EVs were negligible in healthy donors and Stage I breast cancer patients.
• EV fractions increased from Stage II (18%) to Stage III (29%) for EpCAM-positive EVs, with a similar trend for CD24-positive EVs.
• Both markers detected HER2-positive breast cancer at Stages II, III, or IV, differentiating stages except III and IV.

Conclusions:

• AI-assisted DISVT provides a simple, sensitive, and efficient platform for quantitative EV subtype characterization.
• This technology holds significant potential for both basic research and clinical applications in cancer diagnostics.
• The developed method accurately detects and stages HER2-positive breast cancer using EV marker analysis.