Leveraging Spatial Relationships in Microscopic Images for Patient Cancer Diagnosis

Jongwoo Kim, Min Kyu Jeong, Hyeongmin Park, Young Sin Ko, Mun Yong Yi

Abstract: Cancer remains a significant global cause of mortality, underscoring the importance of early detection. While artificial intelligence plays a significant role in transforming the medical industry, no effective solution is available in automatically diagnosing cancer, at the patient level, using microscopic images. In this study, we propose a novel, integrative solution, called Micro Mapper, which leverages the similarity between cell images to comprehend absolute cell positions and their surrounding contexts. Through the consideration of spatial relationships between cells, Micro Mapper enhances the accuracy of cancer prediction at the patient level. Extensive experiments conducted using not only a benchmark dataset but also a real-world dataset show that Micro Mapper is superior over the well-known baseline models. The study findings have direct implications on enhancing microscope-based cancer diagnosis practices.