Speaker: Samiran Dey, Indian Association for the Cultivation of Science, Kolkota
Abstract: Transcriptomic profiling provides rich molecular insights for cancer diagnosis and prognosis, but its high cost limits routine clinical use, where histopathology remains the primary diagnostic modality. Recent advances in artificial intelligence suggest that molecular information can be inferred directly from digital pathology images. This talk discusses a generative multimodal framework that synthesizes transcriptomic features from whole-slide histopathology images and incorporates them to improve cancer grading and survival risk prediction across multiple cancer cohorts. The approach achieves performance comparable to models using real transcriptomic data while relying only on routinely available histopathology images. To address reliability in generative and predictive models, a conformal prediction–based framework is further used to quantify and calibrate uncertainty, providing coverage guarantees and improving fairness across gene categories and demographic subgroups. These results highlight the potential of generative multimodal learning and calibrated uncertainty to enable accurate, interpretable, and reliable AI systems for cancer diagnosis and prognosis.
This will be a free hybrid seminar. To register to attend remotely, please click here:Â https://cam-ac-uk.zoom.us/meeting/register/MSeOkbjoRr2eCc1JQaumTQ