A team of researchers from the University of Maine, led by Ph.D. students Jeremy Juybari in electrical and computer engineering and Josh Hamilton in biomedical engineering, has developed an artificial intelligence system called the Context-Guided Segmentation Network (CGS-Net) that improves the accuracy of breast cancer detection in tissue samples. The AI tool mimics the way human pathologists analyze microscopic images, combining detailed local cell-level information with broader tissue context to enhance diagnostic precision. Breast cancer is the second leading cause of cancer-related deaths in women, with one in eight women affected over their lifetime. Current diagnosis relies on pathologists manually examining chemically stained tissue slides under a microscope—a process that is time-consuming and requires specialized expertise. However, two-thirds of the world’s pathologists are concentrated in just 10 countries, creating significant diagnostic delays in underserved regions. In countries like India, where access to timely diagnostics is limited, up to 70% of cancer deaths are linked to treatable conditions. CGS-Net addresses this challenge by using a dual-encoder deep learning architecture that simulates the way pathologists zoom in and out while examining a slide. One encoder analyzes a high-resolution patch to capture fine cellular details, while the other processes a lower-resolution, broader view of the surrounding tissue to understand architectural patterns. Both views share the same central pixel, ensuring alignment and enabling the model to integrate local and contextual information simultaneously. The system was trained on 383 digitized whole-slide images of lymph node tissue, learning to distinguish between cancerous and non-cancerous regions. Results showed that CGS-Net consistently outperformed traditional single-input AI models in identifying cancerous areas. The researchers attribute the improvement to a novel training algorithm and initialization strategy that better capture the holistic nature of tissue analysis. The findings were published in Scientific Reports, with co-authors including faculty researchers Andre Khalil, Yifeng Zhu, and Chaofan Chen. The team has made both the dataset and source code publicly available to promote transparency and collaboration across the global research community. While the current version focuses on binary segmentation—identifying cancer versus non-cancer—the researchers envision future applications including multiclass tissue analysis, integration with other cancer types, and combining AI with multimodal data such as radiology images or molecular profiles to further improve accuracy. Importantly, the goal is not to replace pathologists but to support them. By enabling faster, more accurate analysis, tools like CGS-Net could help reduce diagnostic delays, especially in areas with limited access to specialists. The project highlights the power of interdisciplinary research, merging engineering, computer science, and biomedical innovation to address pressing global health inequities. As digital pathology grows, AI systems that emulate human expertise offer a promising path toward more accessible, equitable, and life-saving cancer care.