Bone tumor recognition strategy based on object region and context representation in medical decision-making system

Bone tumors are a leading cause of morbidity and mortality in human health. The application of artificial intelligence in medical assistance has fundamentally transformed traditional, labor-intensive diagnostic methods, effectively alleviating the pressure on medical resources. However, the multi-scale nature of bone tumors in medical images, along with complex tumor boundaries and disordered textures, makes it difficult for algorithms to distinguish normal tissue from tumor tissue when relying solely on pixel-level or contextual information for segmentation. To address this, this paper proposes a bone tumor recognition strategy based on object region and context representation (RCROS), which enhances pixel-level features using object region and context representation. The RCROS strategy aggregates pixel representations from each tissue category to estimate the representation of the corresponding object region, and then calculates the relationship between each pixel and its target region. Finally, the object context representation is employed to enhance the representation of each pixel. Experiments were conducted using more than 80,000 datasets from Huaihua Second People's Hospital. RCROS achieves high accuracy while maintaining low resource consumption. It reduces the time doctors spend viewing images and provides a more accurate reference for clinical decision-making.