SUMMARY
The research of Wenchao Han, Ph.D., focuses on advancing clinical translational research in pathology. Dr. Han's work includes developing and applying machine learning algorithms for image analysis for digitized pathological specimens and predictive modeling using laboratory testing data. These innovations enhance pathology workflows and enable more accurate and comprehensive disease characterization, ultimately improving patient management.
Dr. Han's research scope encompasses developing, evaluating, and validating machine learning and image analysis algorithms. Dr. Han applies these algorithms to tasks such as semantic and instance segmentation in tissue image analysis, pathological image classification, whole slide image processing, and descriptive report mining leveraging large language models.
Focus areas
- Supervised, weakly supervised and self-supervised learning. Dr. Han develops algorithms and frameworks for computer vision tasks. He applies them to digitized histopathological images for applications such as prostate cancer detection and grading, invasive tumor front identification in bladder cancer, and breast cancer detection and subtyping. He co-leads three projects focusing on developing multimodality vision-language foundational models for pathology images using self-supervised learning.
- Semantic and instance segmentation for microscopic images. Dr. Han developed DeepCSeg, a novel framework for simultaneous instance segmentation of whole cells and cell nuclei in immunofluorescence multiplexed (MxIF) images. He leads the Quantitative IHC project, where his team developed the first unsupervised algorithm for cell identification and expression determination in Ki-67-stained immunohistochemistry images of breast tissue specimens. He is extending this method to other immunohistochemistry images with different markers and specimens from other organ types.
- Domain adaptation and generalization. Robustness and generalizability are critical challenges in artificial intelligence (AI) and machine learning research, especially for real-world applications. Dr. Han is developing methods to properly evaluate AI models for robustness and generalizability. He also is working on innovative techniques for domain adaptation and generalization to address these challenges.
- Graph theory and graph neural networks. In collaboration with the research group from the University of Toronto, Dr. Han developed a framework using graph theory to quantify spatial cell arrangements and cell-to-cell interactions in MxIF images to study the tumor microenvironment. Also, he employed graph neural networks to automate cell gating in MxIF images. Dr. Han is exploring reliable methods to quantify cell-to-cell interactions and advance tumor microenvironment studies further.
- Whole slide image processing and cloud computing. Whole slide images are high-resolution digital images of tissue specimens. They present significant challenges for large-scale studies due to their enormous file sizes. Dr. Han developed and validated one of the earliest pipelines for whole slide image processing, enabling prostate cancer detection and grading in radical prostatectomy specimens. These specimens are substantially larger than biopsy specimens. His work emphasizes efficient whole slide image processing, and he now focuses on integrating digital imaging and communications in medicine (DICOM) format whole slide images with cloud computing infrastructure to improve scalability.
Significance to patient care
Pathology assessment is the gold standard for diagnosis and prognosis, providing critical guidance to manage patients and treatment decision-making for those patients. Computational tools powered by AI algorithms can make these diagnoses more consistent and accurate and enable the discovery of new biomarkers for improved disease characterization and patient care.
Applying AI-powered tools also can streamline tedious and repetitive tasks such as cell counting. This improves pathology workflow efficiency, reducing operational costs and enhancing healthcare professional and patient satisfaction by shortening processing times.
Professional highlights
- Member, Creative Content Committee, Digital Pathology Association, 2023-present.
- Co-investigator, National Science Foundation Grant Award, 2024.
- Member, Annual Conference Planning Committee, Medical Image Computing and Computer Assisted Intervention Society, 2021.
- Lawson Travel Award, University of Western Ontario, 2018.
- Fellowship awardee, Natural Sciences and Engineering Research Council Computer-Assisted Medical Intervention Training Program, 2015.