Multimodal Entity Coreference for Cervical Dysplasia Diagnosis

摘要: Cervical cancer is the second most common type of for women. Existing screening programs cervical cancer, such as Pap Smear, suffer from low sensitivity. Thus, many patients who are ill not detected in process. Using images cervix an aid has potential to greatly improve sensitivity, and can be especially useful resource-poor regions world. In this paper, we develop a data-driven computer algorithm interpreting based on color texture. We able obtain 74% sensitivity 90% specificity when differentiating high-grade lesions low-grade normal tissue. On same dataset, using tests alone yields 37% 96%, HPV test gives 57% 93% specificity. Furthermore, comprehensive algorithmic framework Multimodal Entity Coreference combining various perform disease classification diagnosis. When integrating multiple tests, adopt information gain gradient-based approaches learning relative weights different tests. our evaluation, present novel that integrates images, Pap, HPV, patient age, which 83.21% 94.79% specificity, statistically significant improvement over any single source alone.