QUANTITATIVE ANALYSIS OF THP1 CELL CONFLUENCY AND PROLIFERATION UNDER TEMPORAL AND PHARMACOLOGICAL CONDITIONS USING DEEP LEARNING TECHNIQUES

Abstract

The proliferation and behavior of THP1 cells, a human monocytic cell line, are critical in understanding various biomedical and pharmaceutical applications. This thesis presents a comprehensive analysis of THP1 cell images categorized into different states: 'D2_PAR30' treated with varying concentrations of the drug (5μg, 20μg, 50μg, and 500μg). The primary objectives are to develop and optimize UNet models for accurate cell segmentation, quantify cell confluency, and analyze cell health based on confluency metrics across these categories. Initially, the THP1 dataset, comprising unique and newly labeled cell images, was preprocessed. Original images (1080x1024) were cropped into smaller sizes (128x128, 256x256, and 512x512) and augmented to enhance dataset diversity. These preprocessed images were then used to train a UNet model for cell segmentation, with the 256x256 dataset yielding the best performance. Hyperparameters, loss functions, batch sizes, and epochs were carefully experimented with to optimize the segmentation accuracy. To optimize the model for edge devices, pruning and quantization techniques were employed. Pruning reduced the model size from 355 MB to 100 MB, while quantization further decreased it to 35 MB, making the model significantly more efficient without compromising accuracy. A pipeline was developed to automate the analysis process. Original cell images were divided into 256x256 segments, each segment's cell confluency and area were predicted, and the results were aggregated to assess the overall confluency and cell area of the original image. This method facilitated the evaluation of cell proliferation and confluency changes over time and under different drug treatments, enabling differentiation between healthy and unhealthy cells based on confluency. The analysis revealed distinct patterns of cell confluency and proliferation associated with temporal changes and drug treatments. By testing 10 images from each category, significant insights were gained into the cellular response under different conditions. These findings contribute to the broader understanding of THP1 cell behavior and provide a foundation for future research in cellular biology and pharmacological studies.