Akademik Veri Yönetim Sistemi
Engineering Applications of Artificial Intelligence, cilt.174, 2026 (SCI-Expanded, Scopus)
Early and precise detection of ovarian neoplasms plays a pivotal role in women’s health, directly influencing patient survival, treatment effectiveness, and overall prognosis. Traditional diagnostic methods, which rely heavily on radiologists’ manual interpretations of computed tomography (CT) scans, often suffer from variability and subjective assessments, resulting in inconsistent diagnostic outcomes. This highlights an urgent necessity to explore innovative computational approaches capable of augmenting human expertise with objective, automated, and highly accurate diagnostic capabilities. Addressing this critical need, we introduce a novel hybrid deep-learning model that seamlessly integrates the strengths of ConvNeXt networks, Convolutional Block Attention Modules (CBAM), and Vision Transformers (ViT). Our proposed ConvNeXt-CBAM-ViT architecture is specifically designed to exploit local and global image features, enhancing the model’s ability to discern subtle pathological differences within ovarian CT images. The incorporation of CBAM attention mechanisms further improves model interpretability by dynamically focusing on diagnostically relevant image regions. The robustness and diagnostic accuracy of our model were extensively validated using three rigorously curated datasets collected by King Abdullah University Hospital (KAUH), Jordan, covering a comprehensive spectrum of ovarian tumor presentations. The datasets — KAUH-OCCTD (Ovarian Cancer CT Dataset), KAUH-BOTD (Benign Ovarian Tumors Dataset), and KAUH-MOTD (Malignant Ovarian Tumors Dataset) — consist of 1943, 1402, and 725 high-quality images, respectively, classified meticulously into normal, benign, and malignant categories, including diverse subtypes. Our model achieved exceptional diagnostic performance, attaining remarkable accuracy scores of 98.05% on the KAUH-BOTD dataset, 98.82% on the KAUH-MOTD dataset, and 98.49% on the comprehensive KAUH-OCCTD dataset. These outcomes underscore the high diagnostic reliability of our hybrid architecture in distinguishing ovarian neoplasms, significantly reducing uncertainty. Ultimately, our innovative approach holds considerable promise to assist clinical decision-making, accelerate diagnostic workflows, and contribute positively toward improved patient outcomes through intelligent and precise CT-based ovarian tumor diagnosis.