Research

Research in the Knowledge Inference in Medical Image Analysis Lab focuses on:

• Unsupervised and self-supervised learning. Although supervised artificial intelligence (AI) is intensively investigated, Dr. Tizhoosh emphasizes the role of nonsupervised topologies in AI. He and his team investigate the design and performance of unsupervised and self-supervised paradigms, such as search, matching, clustering and visualization.
• Generative models in histopathology. Learning histology to generate synthetic histopathology data can be extremely valuable for model development and education. Dr. Tizhoosh and his team investigate the use of many generative adversarial networks in histopathology by looking at clinical data.
• Bias, shortcuts and generalization. Dr. Tizhoosh and his team, and the lab's collaborators, are paying special attention to the generalization capabilities of AI solutions, particularly deep models, to facilitate widespread and reliable deployment of computerized techniques in medicine. Detecting biases and shortcuts in the form of redundant, irrelevant information in data is a major component of most experiments.
• Multimodal and cross-modal patient representation for search and retrieval. Multimodal information retrieval is the major focus of the lab. Dr. Tizhoosh's research culminates in search, retrieval and matching technologies to reduce observer variability. Design and validation of deep topologies that simultaneously learn different patient data and their internal correlations is based on his research in all AI subfields.