Evaluation of Pap Smear Nucleus Cell Image Segmentation: The Impact of Enhancement Processes on Segmentation Result

Abstrak

Accurate nucleus segmentation is vital for automated cervical cancer diagnosis, yet it remains challenging due to overlapping cells and uneven lighting. This study evaluates Polynomial Contrast Enhancement (PCE) using a second-degree polynomial function to improve segmentation on 100 images from the RepomedUNM dataset. The pipeline integrates grayscale conversion, Gaussian blur, and PCE prior to Canny edge detection. Results demonstrate near-perfect Precision (0.9999–1.0000) across all categories (Normal, H-SIL, L-SIL, and Koilocyt), effectively eliminating false positives. However, Recall and Accuracy remained low (max 0.0634 in H-SIL), a technical consequence of Canny’s limitation in capturing thin boundaries versus solid nuclear areas. The study’s novelty lies in the application of second-degree PCE to stabilize intensity variations across multiple diagnostic categories. While PCE ensures exceptional localization precision, future systems should integrate deep learning to enhance recall in complex overlapping structures.

Identifier

• DOI: https://doi.org/10.37034/medinftech.v4i1.93
• OpenAlex: https://openalex.org/W7147267184