Pediatric malignancies are the second leading cause of death in the pediatric population, with solid tumors accounting for approximately 60% of all pediatric malignancies. The pathological diagnosis of pediatric solid tumors is highly complex and specialized, because of its diverse tissue morphology, rare tumor subtypes and lack of labeling data, the traditional pathological diagnosis relies on the experience of senior pathologists, but in actual clinical practice, due to the lack of expert resources and inconsistent diagnostic standards, more efficient and accurate auxiliary diagnostic tools are urgently needed. In this study, we aim to construct a multimodal dataset by collecting high-quality pathological images and pathological diagnosis results of pediatric solid tumors (neuroblastoma, medulloblastoma, Wilms tumor, hepatoblastoma, rhabdomyosarcoma, etc.), and introduce medical knowledge enhancement strategies on this basis, and improve the medical reasoning ability and adaptability to fine-grained pathological tasks by injecting domain knowledge (such as molecular characteristics of tumors, pathological grading standards, diagnostic rules, etc.) into the model. Through the model, the representation space of images and texts is unified, and diversified diagnostic tasks of pediatric solid tumors such as tumor region segmentation, cancer detection, and tumor subtype identification are realized, providing intelligent support for the accurate diagnosis and personalized treatment of pediatric solid tumors.
See this in plain English?
AI-rewrites the medical criteria so a patient or caregiver can understand them. Always confirm with the trial site.
Diagnostic accuracy of patients
Timeframe: immediately after surgery