Predictive Modeling in Cancer Cell Separation: A Machine Learning Approach in DLD Devices

Md Tanbir Sarowar

doi:10.7273/000007237

Deterministic Lateral Displacement (DLD) devices serve as a powerful tool in the field of microfluidics, enabling label-free, size-based separation of particles and cells. These devices offer significant potential for cancer diagnostics, specifically in isolating circulating tumor cells (CTCs) from blood samples to facilitate early detection and improve patient outcomes. Due to the challenge of identifying rare CTCs among the vastly larger population of blood cells, DLD technology optimizes separation through carefully designed geometric configurations, focusing on parameters such as row shift fraction, post size, and gap distance to effectively differentiate cancer cells based on their unique physical properties. This thesis explores how fine-tuning these parameters in DLD devices can lead to more precise and reliable isolation of lung cancer cells, supporting advancements in early cancer diagnostics. In addition to DLD design optimization, this study integrates machine learning models to enhance the process of parameter selection, reducing the reliance on exhaustive simulations and physical prototyping. A large dataset, generated through validated numerical models, underpins the training of various machine learning algorithms, including gradient boosting, k-nearest neighbors (kNN), random forest, and MLP regressor, each tailored to predict particle trajectories and improve separation efficiency. These models are not only instrumental in accurately predicting cell migration patterns within the DLD devices but also serve to identify optimal device configurations rapidly, thus enabling high-throughput and cost-effective cancer cell separation. The application of machine learning in this research extends beyond trajectory prediction; it systematically isolates crucial design parameters essential for advancing DLD technology in cancer research. By analyzing migration characteristics and predicting separation outcomes based on model input, the thesis provides a frame-work for automated DLD device design, offering a streamlined approach for efficient, scalable, and precise cancer cell separation. Ultimately, this predictive modeling approach, combining the strengths of machine learning and microfluidics, is poised to support early cancer detection, thereby contributing to more accessible and targeted therapeutic strategies in precision medicine.

Predictive Modeling in Cancer Cell Separation: A Machine Learning Approach in DLD Devices

Files and links (1)

Abstract

Metrics

Details