Segmentation of Adherent Cell Images

In this international conference paper, we provide a detailed methodology for segmenting adherent cell images, a critical process for advancing our understanding of cellular structures and functions. We begin by establishing foundational background knowledge, including fundamental concepts of cellular architecture and biological functions. Subsequently, we elaborate on our technical approach which integrates advanced image processing algorithms (such as watershed transformation and edge detection) with machine learning techniques (including convolutional neural networks for feature extraction). The implementation utilizes Python's OpenCV and scikit-learn libraries for preprocessing, segmentation, and classification operations. We present comprehensive experimental results with quantitative analysis using metrics like Dice coefficient and Intersection over Union (IoU). The paper further discusses the advantages of our hybrid method in handling overlapping cell boundaries and its limitations in low-contrast scenarios. Through this work, we aim to contribute robust computational tools to cytology and medical research, facilitating improved disease understanding and therapeutic development.