Advancing Pediatric Cancer Diagnosis: Unveiling ecDNA Patterns in FFPE Tissue Images Using Machine Learning

Our study is driven by the critical need for biomarkers that can enhance the detection of residual disease, evaluate therapy response, and predict cancer recurrence. We’re focusing on extrachromosomal DNA (ecDNA), a unique genetic feature characterized by its circularization and amplification of multiple oncogenes. Understanding ecDNA’s role has become crucial in deciphering cancer aggressiveness and resistance to therapies like chemotherapy and radiotherapy. Currently, investigations into ecDNA rely heavily on costly and time-consuming whole genome sequencing. Our aim is to bridge this gap by introducing a cost-effective, sequencing-independent methodology. This approach is particularly valuable, especially when only FFPE tumor slides are available. By leveraging advanced deep neural networks and machine learning techniques, we aim to develop a practical diagnostic solution that can accurately analyze ecDNA without the need for extensive sequencing. This methodology promises to be more accessible and cost-effective, making it deployable even in regions with limited access to advanced molecular testing infrastructure. Accessing the CBTN data allows us to develop child-specific models, which is quite important, knowing that children’s data are limited. Further, providing access to a diverse range of FFPE tumor samples further enhances the impact of our research, facilitating the development and validation of our methodology on a broad scale.