DEEP LEARNING METHODOLOGIES FOR NUCLEI SEGMENTATION AND MITOSIS DETECTION IN HISTOPATHOLOGICAL IMAGES ANALYSIS
Nooshin Nemati, Refik Samet, Emrah Hançer, Serpil Dızbay Sak, Ayça Kırmızı, Zeynep Yıldırım · Kahramanmaraş Sütçü İmam Üniversitesi Mühendislik Bilimleri Dergisi · 2025
Histopathological image analysis is a pivotal area of medical research that leverages deep learning to derive quantitative insights from Hematoxylin and Eosin (H\&E) stained images. This study aims to enhance the analysis of H\&E breast cancer histopathology images by developing deep learning methodologies focused on nuclei and mitosis. Nuclei provide essential information for disease diagnosis, while mitosis is crucial for cancer grading and prognosis prediction. We propose two methodologies: the first segments nuclei using a U-shaped semantic segmentation architecture called CompSegNet; the second detects and classifies mitotic cells through a hybrid approach combining object detection and fuzzy classification algorithms. To evaluate the effectiveness of these methodologies, we introduce two new publicly available datasets: NuSeC (Nuclei Segmentation and Classification) and MiDeSeC (Mitosis Detection, Segmentation, and Classification). These datasets not only validate our methodologies but also provide valuable resources for developing deep learning models in histopathological image analysis.