Segmentation-Based Morphological Profiling of Resistance Cells via Gromov-Wasserstein Distance Matrices
Session Number
CMPS(ai) 24
Advisor(s)
Yogesh Goyal, Northwestern University, Feinberg School of Medicine
Discipline
Computer Science
Start Date
17-4-2025 10:30 AM
End Date
17-4-2025 10:45 AM
Abstract
Understanding and identifying distinct morphological features of cells is crucial for studying cancer progression and treatment response. This project specifically focuses on Pancreatic cancer cells as the cancer remains one of the deadliest malignancies, with a five-year survival rate below 10%. Recently, the FDA approved the drug Sotorasib to treat cancers with the mutation KRAS G12C, which Pancreatic cancer cells contain. However, resistance to Sotorasib and other treatments, including cisplatin, gemcitabine, and trametinib, remains a challenge for scientists in developing treatments for pancreatic cancer.
The project aims to develop a computational pipeline that segments images of pancreatic cancer cells that have been separately treated with the four drugs. Therefore, a deep-learning-based segmentation model, with an accuracy of ~94% across all conditions was built and used. The segmentation model accurately isolates individual cells, allowing for quantitative analysis of morphological features. By applying GW distance to a pair of cells, we compare the morphological distributions of treated colonies of cells on the four treatments to identify phenotypic shifts indicative of resistance.
Initial results suggest distinct morphological signatures for each treatment. This method provides an interpretable framework for predicting whether or not a colony will be resistant to a certain treatment.
Segmentation-Based Morphological Profiling of Resistance Cells via Gromov-Wasserstein Distance Matrices
Understanding and identifying distinct morphological features of cells is crucial for studying cancer progression and treatment response. This project specifically focuses on Pancreatic cancer cells as the cancer remains one of the deadliest malignancies, with a five-year survival rate below 10%. Recently, the FDA approved the drug Sotorasib to treat cancers with the mutation KRAS G12C, which Pancreatic cancer cells contain. However, resistance to Sotorasib and other treatments, including cisplatin, gemcitabine, and trametinib, remains a challenge for scientists in developing treatments for pancreatic cancer.
The project aims to develop a computational pipeline that segments images of pancreatic cancer cells that have been separately treated with the four drugs. Therefore, a deep-learning-based segmentation model, with an accuracy of ~94% across all conditions was built and used. The segmentation model accurately isolates individual cells, allowing for quantitative analysis of morphological features. By applying GW distance to a pair of cells, we compare the morphological distributions of treated colonies of cells on the four treatments to identify phenotypic shifts indicative of resistance.
Initial results suggest distinct morphological signatures for each treatment. This method provides an interpretable framework for predicting whether or not a colony will be resistant to a certain treatment.