Abstract 4155: Development of a virtual Cyclin E1 biomarker using Deep Learning from H&E slides for predicting Cyclin E1 overexpression in gynecological malignancy

Key Points

• The aim is to develop a virtual algorithm for predicting Cyclin E1 protein expression in gynecological cancers using whole-slide images.
• Developed a virtual immunohistochemistry algorithm using deep learning from H&E slides
• Analyzed 60 paired digital slides of high-grade serous ovarian carcinoma and uterine serous carcinoma
• Utilized neural network models to learn features predictive of Cyclin E1 expression
• Evaluated model performance with sensitivity, specificity, and ROC AUC metrics
• Achieved strong concordance between predicted and true Cyclin E1 expression levels
• Provided a scalable alternative to traditional IHC, conserving tissue usage
• Facilitated quantitative assessment of Cyclin E1-positive tumor fractions in the tested cancers