Open Access

ARTICLE

Densely Convolutional BU-NET Framework for Breast Multi-Organ Cancer Nuclei Segmentation through Histopathological Slides and Classification Using Optimized Features

Amjad Rehman¹, Muhammad Mujahid¹, Robertas Damasevicius^2,*, Faten S Alamri³, Tanzila Saba¹

1 Artificial Intelligence & Data Analytics Lab, College of Computer & Information Sciences (CCIS), Prince Sultan University, Riyadh, 11586, Saudi Arabia
2 Centre of Real Time Computer Systems, Kaunas University of Technology, Kaunas, LT-51386, Lithuania
3 Department of Information Systems, College of Computer and Information Sciences, Princess Nourah bint Abdulrahman University, Riyadh, 11671, Saudi Arabia

* Corresponding Author: Robertas Damasevicius. Email:

(This article belongs to the Special Issue: Emerging Artificial Intelligence Technologies and Applications)

Computer Modeling in Engineering & Sciences 2024, 141(3), 2375-2397. https://doi.org/10.32604/cmes.2024.056937

Received 02 August 2024; Accepted 10 October 2024; Issue published 31 October 2024

Abstract

This study aims to develop a computational pathology approach that can properly detect and distinguish histology nuclei. This is crucial for histopathological image analysis, as it involves segmenting cell nuclei. However, challenges exist, such as determining the boundary region of normal and deformed nuclei and identifying small, irregular nuclei structures. Deep learning approaches are currently dominant in digital pathology for nucleus recognition and classification, but their complex features limit their practical use in clinical settings. The existing studies have limited accuracy, significant processing costs, and a lack of resilience and generalizability across diverse datasets. We proposed the densely convolutional Breast U-shaped Network (BU-NET) framework to overcome the mentioned issues. The study employs BU-NET’s spatial and channel attention methods to enhance segmentation processes. The inclusion of residual blocks and skip connections in the BU-NEt architecture enhances the process of extracting features and reconstructing the output. This enhances the robustness of training and convergence processes by reducing the occurrence of vanishing gradients. The primary objective of BU-NEt is to enhance the model’s capacity to acquire and analyze more intricate features, all the while preserving an efficient working representation. The BU-NET experiments demonstrate that the framework achieved 88.7% average accuracy, 88.8% F1 score for Multi-Organ Nuclei Segmentation Challenge (MoNuSeg), and 91.2% average accuracy, 91.8% average F1 for the triple-negative breast cancer (TNBC) dataset. The framework also achieved 93.92 Area under the ROC Curve (AUC) for TNBC. The results demonstrated that the technology surpasses existing techniques in terms of accuracy and effectiveness in segmentation. Furthermore, it showcases the ability to withstand and recover from different tissue types and diseases, indicating possible uses in medical treatments. The research evaluated the efficacy of the proposed method on diverse histopathological imaging datasets, including cancer cells from many organs. The densely connected U-NEt technology offers a promising approach for automating and precisely segmenting cancer cells on histopathology slides, hence assisting pathologists in improving cancer diagnosis and treatment outcomes.

---

Keywords

---