Microfluidics-based patient-derived disease detection tool for deep learning-assisted precision medicine

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Oncology

Background:

• Cancer's heterogeneity complicates prognosis and treatment, driving the need for advanced diagnostic tools.
• Current methods for cancer assessment can be time-consuming and require significant manual input.
• Liquid biopsies offer a minimally invasive method for cancer detection and monitoring.

Purpose of the Study:

• To develop and validate an Intelligent Disease Detection Tool (IDDT) for routine cancer prognosis and treatment assessment.
• To integrate a microfluidic tumor model with deep learning for automated analysis of liquid biopsies.
• To create a patient-centric, cost-effective tool for precise clinical decision-making.

Main Methods:

• Development of a microfluidic-based tumor model for high-throughput 3D tumor culturing.
• Integration of an advanced deep neural network (Mask R-CNN, vision transformer, SAM) for automated analysis and segmentation.
• Clinical validation using liquid blood biopsy samples from 71 cancer patients and healthy donors.

Main Results:

• The IDDT achieved high performance with a mean Intersection Over Union (mIoU) of 0.902 and reduced manual labeling time by up to 90%.
• Rapid image analysis (<2s per image) enabled efficient clinical cohort classification.
• The tool accurately stratified healthy donors (n=12) and cancer patients (n=55) with high precision (99.3%) and sensitivity (98%).

Conclusions:

• The IDDT provides an intelligent, label-free, and cost-effective solution for cancer detection and prognosis.
• This technology facilitates precise medical decisions and personalized treatment strategies for cancer patients.
• The microfluidic-based approach combined with deep learning shows significant potential for routine clinical application in oncology.