IMPROVED CELL SEGMENTATION USING UNET

ABSTRACT

This project aims to enhance cell segmentation using the UNET architecture, a crucial component of biomedical image analysis for tasks like cell quantification and disease detection. While UNET has shown potential in this regard, its demanding computational requirements pose a significant hurdle. The primary purpose of this work is to optimize computational efficiency while addressing the complexities inherent in cell segmentation tasks, including diverse cell types, shapes, and staining techniques. The solution employed in this project leverages the U-Net network alongside innovative image pre-processing techniques for effective segmentation. While a pre-processing method was introduced to address the challenge posed by large annotations, it was only partially successful, prompting the creation of a tailored training dataset through selective data extraction. The outcomes are encouraging, demonstrating that when combined with data enhancement strategies, U-Net mitigates computational challenges, yielding commendable segmentations, albeit with residual inaccuracies arising from false positives.