Diagnosis of Blood Diseases and Disorders with Topological Deep Learning

Blood diseases and disorders, including leukemia and infectious diseases of red blood cells, pose significant diagnostic challenges due to their complex presentations and reliance on time-consuming cytomorphological analysis to detect subtle morphologic features. While microscopic examination remains the gold standard in their diagnosis, its dependence on expert interpretation high-lights the need for advanced methods to enhance the diagnostic workflow. Recently, deep learning (DL) methods have shown promise in medical imaging by automating and improving accuracy. However, these approaches often require large, annotated datasets, which are scarce for rare diseases, and they often face interpretability issues, limiting their integration into clinical practice. In this study, we present a novel framework that integrates topological features with state-of-the-art DL techniques to enhance the analysis of cytomorphological images for diagnosing blood disorders. By combining global topological information with the localized patterns captured by DL models, our approach improves diagnostic accuracy while ensuring robustness, interpretability, and reproducibility. Experimental results demonstrate that the inclusion of topological features not only enhances model performance but also proves particularly effective in limited data settings. This methodology addresses critical limitations of existing techniques, advancing the classification and diagnosis of blood disorders while improving efficiency and reliability in medical imaging workflows.