中文版本

Opportunity

Cancer remains a leading cause of death globally, with conventional diagnostic methods relying on invasive solid tumor biopsies that are uncomfortable, time-consuming, and cumbersome. Liquid biopsies offer a minimally invasive alternative for detecting disease-related biomarkers, but current approaches face significant limitations. Detection of protein or gene biomarkers often requires targeted probe labeling, which depends on prior knowledge of biomarker characteristics and struggles with tumor heterogeneity and phenotypic changes during treatment. Traditional cancer research frequently uses commercially available cell lines that lack clinical relevance and fail to replicate the complex tumor microenvironment, including interactions with non-cancerous cells. Circulating tumor cells (CTCs) from patient blood better reflect tumor heterogeneity and hold promise as biomarkers, but they are rare in liquid biopsies (typically 0–1000 CTCs per mL of blood) and prone to apoptosis after shedding. Existing CTC expansion techniques suffer from long culture cycles (over six months), low efficiency (below 20%), and high costs, while most current algorithms focus on non-clinical spheroid characterizations and require extensive manual annotation, making them impractical for timely clinical decision-making. There is a pressing need for an improved, fully automated, and unique prediction platform that can provide early, accurate prognostic insights with minimal invasiveness and rapid turnaround.

Technology

This invention presents an integrated, patient-derived liquid biopsy-based platform (LIQBP) designed for early prediction of disease prognosis and evaluation of treatment efficacy. The platform comprises three main components: an in vitro disease model, a cell cluster image processing and analysis tool, and an image acquisition module. The in vitro disease model utilizes a microfluidic device, typically a biochip with at least two layers—a bottom layer featuring multiple elliptical microwells for cell cluster formation and a top barrier layer to prevent fluid mixing between wells. Cells from a patient’s liquid biopsy (e.g., peripheral blood) are cultured in these microwells to form clusters within approximately 14 days, without requiring any labeling agents. The image acquisition module, often a phase-contrast microscope, captures images of these clusters in formats such as bright-field, dark-field, differential interference contrast, or phase-contrast. The core innovation lies in the fully automated analysis tool, which employs deep learning neural networks (e.g., ResUnet++ for image segmentation and Resnet 34 for classification) to process these label-free images. The system automatically performs flat-field correction, ellipse detection, edge detection, and morphological characterization. It extracts and quantifies key morphological parameters from the cell clusters—size, thickness, roughness, and compactness—derived from features like normalized gray value (nGV), normalized standard deviation of gray value (nSD^GV), and ratios such as RGVS (nGV to cluster size) and RGVSD (nGV to nSD^GV). These parameters enable the distinction between healthy individuals and cancer patients, prediction of cancer stages (including TNM staging), and assessment of treatment response across different therapy cycles, all within a short timeframe (as quick as one minute per batch analysis).

Advantages

• Provides a non-invasive, label-free prognostic tool using patient-derived liquid biopsies, eliminating the need for dyes or probes and associated toxicity. 
• Enables rapid, high-throughput analysis with results potentially available within minutes, significantly reducing labor and accelerating treatment decisions. Offers high sensitivity and specificity in stratifying clinical cohorts, distinguishing healthy donors from cancer patients and predicting cancer stages. 
  
• Integrates deep learning for fully automated image segmentation and classification, minimizing human error and subjective interpretation. 
• Utilizes cost-effective microfluidic technology and standard microscopy, making it accessible even in resource-limited settings. 
  
• Reflects disease heterogeneity across cancer types through quantitative morphological parameters derived directly from cell clusters. 
  
• Supports personalized medicine by allowing evaluation of treatment efficacy and optimization of therapeutic strategies for individual patients. 
  
• The platform is flexible and can be customized with additional functionalities or adapted for various applications.
  

Applications

• Early prediction and monitoring of cancer prognosis in clinical oncology. 
• Evaluation of treatment efficacy across different therapy cycles (e.g., chemotherapy, immunotherapy). 
• Stratification of cancer patients based on TNM staging or overall cancer stage. 
• Distinction between healthy individuals and those with malignant conditions using liquid biopsies. 
• Drug discovery and validation, including screening of new drug combinations in a patient-specific context. 
• Point-of-care diagnostics in decentralized healthcare settings, potentially with portable imaging systems. 
• Research applications in studying tumor biology, heterogeneity, and microenvironment interactions. 
• Integration into clinical workflows for routine prognostic assessment and personalized treatment planning.